An Interference Cancellation Scheme for Mobile Communication Radio Repeaters

Transcription

1 1778 IEICE TRANS. COMMUN., VOL.E92 B, NO.5 MAY 2009 PAPER Special Section on Radio Access Techniques for 3G Evolution An Interference Cancellation Scheme for Mobile Communication Radio Repeaters Moohong LEE a), Student Member, Byungjik KEUM, Young Serk SHIM, and Hwang Soo LEE, Nonmembers SUMMARY An interference cancellation (ICAN) scheme for mobile communication radio repeaters is presented. When a radio repeater has a gain that is larger than the isolation between its transmit and receive antennas, it oscillates due to feedback interference signals. To prevent feedback oscillation of a radio repeater, we first formulate a feedback oscillation model of the radio repeater and then derive an ICAN model from that model. From the derived ICAN model, we show that the stability and the signal quality of the repeater depend on the repeater s gain and delay, the propagation delay on feedback paths, feedback channel characteristics, and the capability of the feedback channel estimation algorithm. It is also shown that the stability condition of the repeater does not guarantee the quality of the repeater s output signal. To guarantee repeater s stability and signal quality, an ICAN scheme based on an iterative algorithm is subsequently proposed. The simulation results confirm the relationship between the stability and signal quality of the repeater and the impact of the aforementioned factors. Using the proposed ICAN scheme, a mean error vector magnitude (quality indicator) of about 6.3% for the repeater s output signal was achieved. key words: interference cancellation, repeater, feedback, stability, signal quality 1. Introduction A radio repeater that uses the same frequency for transmit and receive signals can go into feedback oscillation regardless of the input signal when the repeater s gain is larger than the isolation between its transmit and receive antennas. Such feedback oscillation prevents the use of the repeater s maximum available output power [1] [5] and causes the repeater to transmit unwanted spurious signals into the air [2]. In addition, it makes installation of the repeater difficult, because sufficient isolation should be secured between the transmit and receive antennas of the repeater [2]. The isolation between the two antennas can be improved by controlling the antenna patterns and pointing direction, and by reducing leakage from the repeater s body, the antenna feed cables, and the joints of the cables and the connectors. In addition, unwanted antenna coupling effects may be reduced by physical separation of the two antennas and by introducing shielding between the two antennas [2]. However, such efforts cannot prevent the feedback signals from entering the receive antenna of the repeater, because they are generated by reflection of the repeater s output sig- Manuscript received August 25, Manuscript revised November 18, The authors are with the School of EECS, KAIST, Daejeon Korea. Part of this manuscript was first published in the 4th International Conference on Wireless and Mobile Communication, a) wildgoosemh@mmpc.kaist.ac.kr DOI: /transcom.E92.B.1778 nal from objects such as buildings and mountains around the repeater. Accordingly, feedback interference cancellation (ICAN) techniques [6] [9] have been developed to cancel the feedback signals. Among them, analog ICAN techniques first estimate the amplitude, phase, and delay of feedback signals and then cancel the feedback signals using generated signals with the same amplitude and anti-phase as the feedback signals [4], [5]. However, accurate estimation of the amplitude, phase, and delay for each of the multiple feedback delayed signals in the analog domain presents a challenge when the feedback channel is changing over time. On the other hand, digital ICAN techniques [6] [9] are based on an adaptive filter. They have emphasized aspects of implementation such as cancellation of multiband feedback signals [6], computational complexity reduction [7], and an efficient simulation method [8]. An ICAN model for a radio repeater was briefly introduced in [9] as a part of our work. In this paper, we propose an ICAN scheme based on an ICAN model to characterize feedback interference signals and cancel them in mobile communication radio repeaters. Applying the proposed ICAN scheme, we show that the stability and signal quality of a radio repeater depend on the repeater s gain and delay, the propagation delay on feedback paths, feedback channel characteristics, and the capability of the feedback channel estimation algorithm. The ICAN repeater is briefly described in Sect. 2. In Sect. 3, a feedback oscillation model and an ICAN model for a radio repeater are formulated. In Sect. 4, the proposed ICAN scheme based on the ICAN model is introduced. The stability and signal quality of the repeater are also presented. Simulation results that demonstrate the validity of the proposed ICAN scheme are presented in Sect. 5. Finally, conclusions are given in Sect The Interference Cancellation Repeater 2.1 Structure The structure of the downlink path for a radio ICAN repeater, which relays a signal from a base station (BS) to mobile stations (MSs), is shown in Fig. 1. It is composed of a radio frequency (RF) receiver, an analog-to-digital converter (ADC), a feedback ICAN system (FICANS), a digital-toanalog converter (DAC), and a RF transmitter. The RF receiver performs pre-processing such as amplification, frequency downconversion to an intermediate frequency (IF), Copyright c 2009 The Institute of Electronics, Information and Communication Engineers

2 LEE et al.: AN INTERFERENCE CANCELLATION SCHEME FOR MOBILE COMMUNICATION RADIO REPEATERS 1779 Fig. 1 The structure of a feedback ICAN repeater. and filtering for the incoming analog signal. The ADC converts the IF analog signal to a digital signal. Feedback ICAN is performed by the FICANS in the digital domain. The DAC again transforms the interference cancelled digital signal to an analog signal. The RF transmitter executes post-processing, which includes filtering, frequency upconversion to RF frequency, and power amplification to radiate the RF signal into the air [6] [8]. Parts of the output signal radiating from the repeater s transmit antenna toward MSs are reflected from reflectors around the repeater and enter the receive antenna of the repeater. This feedback signal is combined at the receive antenna with the input signal coming from the BS. If the repeater s gain is larger than the isolation between the transmit and receive antennas of the repeater, the amplitude of the feedback signal continues to increase every time the signal goes through a feedback closed loop (FCL) composed of the repeater and the feedback channel. This phenomenon is called feedback oscillation (FOSC). Eventually, the repeater becomes saturated due to FOSC. However, if a FICANS in the repeater cancels the feedback signal every time it enters the repeater, the repeater will not go into FOSC [9]. In general, the feedback ICAN operation is performed at a low frequency range in order to obtain more accurate estimation and cancellation of signals. For this purpose, the first frequency conversion is executed in the RF receiver before the ADC and the second frequency conversion in the FI- CANS after the ADC. However, this frequency conversion process does not affect the FOSC of the repeater, because it does not limit the increase of the signal level due to FOSC [9]. The purpose of a radio repeater is to receive a small input signal with a receive antenna, to amplify it without distortion, and to transmit it via a transmit antenna. Hence, the output signal of the repeater should always meet the spectrum mask specifications [10]. This requires that the receive antenna of the repeater be aligned to the antenna of the BS for high quality signal reception with a high signal to noise ratio (SNR), that is, SNR > 30 db. Therefore, the average power of the input signal to a radio repeater is almost constant. In addition, in order to obtain a large coverage with the small input power, the maximum gain G m of the repeater should be used. When a line of sight between the receive antenna of the repeater and the antenna of the BS is not maintained, the average power of the input signal to a radio repeater slowly changes over time. In this case, an automatic level control function is turned on so as to maintain the average output power at a constant level in spite of slow input power variation. This causes the gain of the repeater to change over time. Thus, the isolation between the transmitter and receiver antennas of the repeater should be sufficiently secured so that the isolation is always larger than the maximum gain G m of the repeater, with some margin to account for the variability of the channel [1], [3]. Otherwise, the repeater will oscillate and soon become saturated. In this work, the repeater s gain G 0, instead of the repeater s maximum gain G m, is used for convenience of description and analysis. 2.2 Feedback Oscillation Conditions Basically, the FOSC of a radio repeater occurs when the amplitude of the feedback signal at a component frequency within the given bandwidth is two times larger than that of the input signal to the repeater regardless of the phase condition between the feedback signal and the input signal. In other words, when the repeater s gain G 0 is two times larger than the isolation L ISO between the transmitter and the receiver of the repeater, or when the repeater s equivalent gain G, which is defined as G 0 /L ISO, is larger than two (G > 2), the radio repeater oscillates. In the case that 1 G 2, the radio repeater may oscillate or not oscillate depending on the phase condition between the feedback signal and the input signal. In real repeater systems, the phases of the feedback signal and the input signal are almost random due to random time delay variation. Therefore, if the condition where 1 G 2 lasts for a long time, the possibility that the radio repeater oscillates will be high, because there are more chances that the feedback signal and the input signal will be in phase or partially in phase. In general, the probability of FOSC of a radio repeater whose equivalent gain is in a range where 1 G 2 becomes higher if G is higher, the repeater s bandwidth is wider, or the condition 1 G 2 lasts longer. Accordingly, even though there is no input signal to the radio repeater from the BS, the radio repeater will oscillate due to the thermal noise that exists on the repeater if the condition G > 1 is met and it lasts for a long time. Of course, the continual input signal, which comes from the BS and is considerably large compared to the noise existing on the repeater, will cause the radio repeater to be saturated more quickly compared to when there is no input signal to the repeater, if the oscillation condition mentioned above is met. 3. Interference Cancellation Model 3.1 A Feedback Oscillation Model The analog signal, which propagates through a feedback channel and is processed in the RF receiver of a radio repeater, as shown in Fig. 1, may be represented in the digital domain without any loss of information through an analogto-digital conversion process. Hence, the FOSC that occurs

3 1780 IEICE TRANS. COMMUN., VOL.E92 B, NO.5 MAY 2009 x(n) = G[i(n d) + y(n d)]. (2) If y(n) in (1) is applied to (2), x(n) becomes x(n) = Gi(n d) + Gh(n d) x(n d t ) (3) where the total delay d t = k 0 + d. Since it is assumed that the impulse response h(n) and the delay k 0 do not change over time, after x(n) in(3)is applied to itself twice, the output signal x(n) at time n = 3d t + d can be expressed by Fig. 2 A simple model of feedback oscillation (with SW connected to the P1) and an ICAN model (with SW connected to the P2) based on an estimated channel filter. in a FCL can be described in the digital domain by the simple model shown in Fig. 2, where the switch SW is connected to port P1. A normal radio repeater amplifies the small input signal that enters the input antenna, and transmits the amplified output signal after some time delay via the transmit antenna. Therefore, the repeater can simply be represented by the repeater s equivalent gain G and time delay d, as shown in Fig. 2. Portions of the output signal radiating from the repeater s transmit antenna toward MSs are reflected from reflectors around the repeater and arrive at the receive antenna of the repeatervia severalfeedbackpaths with different time delays. Hence, the feedback channel can be modeled by a channel filter with an impulse response h(n) and a delay block with a delay k 0. The impulse response h(n) is normalized to the dominant feedback component with the maximum power so that h(n) 1foranytimen. The delay k 0 in the delay block indicates the propagation delay of the dominant feedback component. Generally, the transmitter and receiver antennas of the repeater are tightly fixed against outdoor environmental factors such as wind, rain, and snow for stable transmission and reception of the signal at a specified direction. Furthermore, reflectors such as mountains and tall buildings around the repeater, which generate feedback signals, can normally be assumed not to move over time, although slight movement caused by wind or earthquakes for a short time is possible. Therefore, in this work, it is assumed that the impulse response h(n) and the delay k 0 do not change over time. The input signal i(n) in Fig. 2 represents a signal that comes from a BS or the sum of a signal that comes from a BS and noise that is generated in the repeater itself. The feedback signal y(n) can be expressed by y(n) = h(n) x(n k 0 ) = L h( j)x(n k 0 j) (1) where * denotes the convolution and L indicates the length of the channel impulse response h(n). In (1), the output signal x(n) is x(3d t + d) = Gi(3d t ) + G 2 h(3d t ) i(2d t ) + G 3 h(3d t ) h(2d t ) i(d t ) + G 4 h(3d t ) h(2d t ) h(d t ) i(0). (4) In (4), it is assumed that x(d) = Gi(0). The average power of the input signal coming from the BS is almost constant and the noise process existing on the repeater has constant variance. In addition, they are uncorrelated. Thus it can be assumed that the average power of the input signal i(n) is constant. On the other hand, FOSC of a radio repeater occurs as a result of the increasing power of the input signal i(n) that circulates in a FCL, not by the type of information in the input signal i(n). Hence, from the viewpoint of FOSC, the input signal i(n) can be regarded as a time invariant signal. As a result, using the commutative property of convolution [11], the following relation can be obtained, h(n) i(n k 0 ) = h(n k 0 ) i(n). (5) If (5) is applied to (4), the output signal x(n) at time n = 3d t + d is x(3d t + d) = Gi(3d t ) + G[Gh(2d t ) + G 2 h(2d t ) h(d t ) + G 3 h(2d t ) h(d t ) h(0)] i(3d t ). (6) Using the same logic, the output signal x(n) at time n = Md t + d can be represented by x(md t + d) = Gi(Md t ) M m +G {Gh(Md t jd t )} i(md t) (7) m=1 where M is an integer. In (7), the symbol Π denotes that a series of convolution operations is performed on delayed impulse responses. For example, 2 h(mk 0 jk 0 ) = h(mk 0 k 0 ) h(mk 0 2k 0 ). (8) In (7), the first term after the equal sign is the normal output signal of the repeater and the remaining terms are generated by the feedback of the repeater s output signal via the feedback channel. When G < 1, which means G 0 < L ISO, Gh(n jd t ) < 1foranyn. As a result, x(md t + d) in (7) converges to Gi(Md t ) + α, α being a finite value, as M increases. When G > 1, which means G 0 > L ISO, Gh(n jd t ) > 1forsome

4 LEE et al.: AN INTERFERENCE CANCELLATION SCHEME FOR MOBILE COMMUNICATION RADIO REPEATERS 1781 n. As a result, x(md t + d) in (7) will continue to increase as M increases. This ever-increasing output signal level of the repeater causes the repeater to go into FOSC, resulting in saturation of the repeater for a large M. 3.2 Feedback Interference Cancellation Model In order to prevent the repeater from going into FOSC when G > 1, the feedback ICAN model (with the SW connected to the P2) illustrated in Fig. 2 can be used. To cancel the feedback signal y(n) whenever it enters the repeater, the estimated feedback signal y E (n) is generated by an estimated channel filter with an impulse response w(n) as follows: y E (n) = w(n) x(n l 0 ) (9) where l 0 is the estimated delay of the dominant feedback signal. The error signal e(n) in Fig. 2 is given by e(n) = i(n) + y(n) y E (n). (10) If (1) and (9) are substituted into (10) and the estimated delay l 0 is assumed to be equal to the propagation delay k 0, e(n) = i(n) + [h(n) w(n)] x(n k 0 ). (11) Therefore, the output signal x(n) can be expressed by x(n)=gi(n d)+g[h(n d) w(n d)] x(n d t ) (12) where the relation x(n) = Ge(n d) is used. By comparing (3) and (12), and exploiting (7), the output signal x(n) at time n = Md t + d can be represented by x(md t + d) = Gi(Md t ) M m +G D(Md t jd t ) i(md t). (13) m=1 In (13), D(Md t jd t ) = G[h(Md t jd t ) w(md t jd t )], and w(n) is assumed to be constant for concise formulation even if they may continue to change slightly due to the ICAN algorithm introduced later in this work. When D(Md t jd t ) > 1, which means G > 1owingto the normalization of h(n), x(md t +d) in (13) continues to increase as M increases, resulting in FOSC. If D(Md t jd t ) = 0, x(md t + d) = Gi(Md t ), which is an ideal case with perfect ICAN. When D(Md t jd t ) < 1, x(md t + d) converges to Gi(Md t )+β, β being a finite value, as M increases. Therefore, to prevent FOSC of the repeater when G > 1, estimation of the feedback channel should be perfect so that h(n) = w(n) foranyn, or it should be accurate enough to obtain the relation D(n jd t ) < 1foranyn. 4. Interference Cancellation Scheme 4.1 Stability and Signal Quality The ICAN model in (13) indicates that the output signal of the radio repeater x(n) is composed of two parts. The first part, which is the first term after the equal sign, is the desired output signal Gi(Md t ). The second part is the remaining terms, which are unwanted interference signals. If these interference signals increase beyond the control range, the repeater goes into FOSC and soon becomes saturated. The second part, which determines the stability of the radio repeater, depends on G, d t, h(n), w(n), and the finite input signal i(n). First, if the repeater s equivalent gain G is large, it is difficult to meet D(n jd t ) < 1 with a fairly accurate channel estimation method. Second, when the total delay d t is small, the number of terms generated by the feedback of the output signal increases further for a given period of time. This means the feedback signal circulates the FCL faster, resulting in a rapid increase of the output signal. Third, if the feedback channel has many multipaths with different power levels and a wide delay spread, accurate feedback channel estimation becomes difficult. Thus, it is more difficult to meet D(n jd t ) < 1foranyn when G > 1. Fourth, inaccurate estimation of d t can also make it difficult to meet the condition D(n jd t ) < 1. Furthermore, the interference signals in (13) make the output SNR worse compared to the input SNR. The bad output SNR results in poor quality of the repeater s output signal. If the input SNR is E{i 2 (n)}/n i, the output SNR can be represented by P o N o = G o E{i 2 (n)} 2 M m G o N i + G o E D(n jd t ). (14) i(n) m=1 In (14), E{} denotes the ensemble average and G o is the repeater s gain. E{i 2 (n)}, P o, N i, N o are the input and output signal powers, and the input and output noise powers, respectively. The noise effect is neglected in (14) because the input SNR is normally high (> 30 db) for high quality relay service. According to (14), even if the condition D(n jd t ) < 1 when G > 1 can prevent FOSC of the repeater, it does not guarantee the quality of the interference cancelled output signal [10]. The smaller the D(n jd t ) is, the better the output SNR is. Therefore, both the stability of the repeater and the quality of the repeater s output signal can be secured only when D(n jd t ) 1foranyn so as to meet the error vector magnitude (EVM) requirement of 12.5% for the repeaters [10]. The EVM of less than 12.5% roughly corresponds to the output SNR of greater than 18 db in (14) according to the relationship between SNR and EVM, i.e. SNR 1/EVM 2 [12]. In that case, (13) becomes x(n) Gi(n d). The signal quality of a repeater is normally measured by the following EVM formula [10], P 1 e k 2 k=0 EVM = 100%. (15) P 1 R k 2 k=0 In (15), e k is the error vector between the kth quadrature

5 1782 IEICE TRANS. COMMUN., VOL.E92 B, NO.5 MAY 2009 Fig. 3 Proposed ICAN scheme. To guarantee both the stability of the repeater and the quality of the repeater s output signal, we propose an ICAN scheme based on the proposed ICAN model in Fig. 2, where an iterative algorithm such as the LMS or RLS algorithm is applied to an adaptive filter to adaptively estimate the changing feedback channel response and cancel the feedback interference signal using the estimated feedback signal, as shown in Fig. 3. From (1) and (9), the condition D(n jd t ) 1for agiveng is equivalent to the condition y(n) y E (n) 1. In real systems, it is difficult to accurately estimate the feedback channel h(n) and the delay k 0 at any n due to the randomness of a wireless channel. Therefore, it is more reasonable to minimize the difference y(n) y E (n) by minimizing the mean square difference E{[y(n) y E (n)] 2 } than by minimizing the difference y(n) y E (n) through estimation of the feedback channel h(n) atanyn. The minimum mean square of the error in (10) is E{e 2 (n)} = E{i(n) 2 } + E{[y(n) y E (n)] 2 } 2E{i(n)[y(n) y E (n)]}. (16) Since the feedback signal y(n) is related to a series of delayed input signals i(n md t ), with m being an integer, the input signal i(n) and the feedback signal y(n) may be considered uncorrelated due to the time difference md t. As a result, E{i(n)y(n)} = 0. For the same reason, the input signal i(n) and the estimated feedback signal y E (n) are uncorrelated, and thus E{i(n)y E (n)} = 0. When these two results are applied to (16), (16) becomes E{e 2 (n)} = E{i(n) 2 } + E{[y(n) y E (n)] 2 }. (17) phase shift keying (QPSK) symbol vector recovered in the receiver and the kth reference QPSK symbol vector R k generated in the signal generator. P is the number of QPSK symbol vectors needed to obtain a reliable EVM. The size of G that can be supported without FOSC depends on the signal quality requirement and the feedback channel estimation accuracy. If G is large, it is more difficult to make D(n jd t ) small with a given channel estimation accuracy. As a result, the likelihood of FOSC becomes higher and the output SNR becomes more deteriorated. Hence, the ICAN scheme should be able to cancel feedback signals accurately enough to guarantee both the stability of the repeater and the quality of the repeater s output signal. For that purpose, the ICAN scheme must have a capability to follow the fast-increasing feedback signals that are generated by a combination of G and d t. It should also be able to accurately estimate multipath feedback channels with different power levels and delay spreads. Furthermore, it should not be sensitive to the estimation accuracy of the propagation delay k 0. Lastly, the computational complexity of the ICAN scheme should be relatively low, because feedback ICAN requires real-time processing. 4.2 Interference Cancellation Scheme Since the average power of the input signal E{i 2 (n)} is assumed to be constant, minimizing E{e 2 (n)} is equivalent to minimizing E{[y(n) y E (n)] 2 }. Therefore, to meet the condition D(n jd t ) 1whenG > 1, conventional iterative algorithms that minimize the mean square error E{e 2 (n)} can be used. 4.3 Interference Cancellation Algorithm Many iterative algorithms that minimize the mean square of the error e(n) are available, including the least mean square (LMS) algorithm, the normalized LMS algorithm, and the recursive least squares (RLS) algorithm [13], [14]. The iterative algorithm used in the feedback ICAN scheme should be able to cancel the fast-increasing feedback signal that is determined by G and d t in real time. Accordingly, the complexity and the convergence rate of the algorithm should be considered at the same time when selecting the iterative algorithm. In addition, the sampling rate of the ADC should also be taken into account, since it determines the iteration number of the algorithm within a given period of time. As an example, consider a case where G = 1dB(G 0 = 80 db and L ISO = 79 db), d t = 4 μs(d = 3 μs andk 0 = 1 μs), and the repeater s input power P i, the maximum output power P max, and the noise power N i are 50 dbm, 40 dbm, and 105 dbm within 10 MHz bandwidth, respectively. A noise figure of 8 db is assumed here. The feedback signal circulates in a FCL once every 4 μs and the power level increase ΔP is 1 db per feedback cycle. The number of feedback cycles K until the repeater s output power P o reaches the maximum level P max is given by K = (P max P out )/ΔP. P o = G 0 + P i in dbs when there is an input signal from absorp o = G 0 + N i in dbs when there is only noise to the repeater. Hence, the repeater begins to be saturated by FOSC after the time T = K d t from the moment a FCL is formed. T 40 μs when there is an input signal from a BS or T 260 μs when there is only noise. If the sampling rate is 50 MHz, there are 200 data samples available per feedback cycle. In general, the LMS and RLS algorithms require on the

6 LEE et al.: AN INTERFERENCE CANCELLATION SCHEME FOR MOBILE COMMUNICATION RADIO REPEATERS 1783 order of N operations and on the order of N 2 operations, respectively, where N is the number of filter taps [13]. The iteration number required for the algorithm to converge to a steady state is about 250 for the LMS algorithm and about 40 for the RLS algorithm [15], although this depends on control parameters such as δ and λ (RLS algorithm), or μ (LMS algorithm). Considering the complexity and the convergence rate, the LMS algorithm is a good candidate for feedback ICAN if it meets the signal quality requirement of the repeater [10]. For better signal quality, the RLS algorithm is preferred, although the complexity is higher. In this work, application of the LMS algorithm to the feedback ICAN scheme for a radio repeater is briefly described. The ICAN scheme in Fig. 3 is based on an adaptive filter that consists of an adaptive channel tap updater (ACTU), an estimated channel filter, a delay block, and an adder block. The ACTU in the adaptive filter calculates the coefficient vector w(n + 1) with a length of N using the error signal e(n), the delayed output signal vector x(n l 0 ), and the previous coefficient vector w(n) as follows: w(n + 1) = w(n) + μe(n)x(n l 0 ). (18) In (18), μ is a parameter to control the convergence rate and the excess mean square error of the LMS algorithm and l 0 is the delay of the dominant feedback signal. The error e(n) is given by (10). The estimated channel filter with the coefficient vector w(n) provided by the ACTU generates the estimated feedback signal y E (n), y E (n) = w T (n)x(n l 0 ). (19) In (19), w T (n) indicates the transpose of w(n). The feedback signal y(n) is cancelled by y E (n) in the adder block, as shown in Fig. 3. The control parameter μ can be optimized for the stability and signal quality of the repeater. The filter length N should be long enough to cover the delay spread of the feedback channel h(n) plus a margin needed to cover inaccurate estimation of the delay k 0. At the initial stage, the LMS algorithm slowly follows the fast-increasing feedback signal, because it starts with a zero coefficient vector. Hence, a limiter, which limits the level of the feedback signals to a specified value, is required to restrict increase of the feedback signals until the LMS algorithm begins to sufficiently cancelthem at the initial stage. The limiter may also prevent a sudden increase of the feedback signal, which is caused by the slowness of the LMS algorithm when an abrupt channel change occurs due to brief feedback path blocking by a flying object or other causes. It may be placed directly after the adder block in Fig. 3. The delay l 0 in (19) is critical for the exact cancellation of the feedback signals and should be equal to the delay k 0 in (1). A simple method of accurately estimating k 0 is to transmit a test signal via the transmit antenna and to measure its delayed version at the receive antenna, as shown in Fig. 3. This delay measurement is performed only once at the initial setup time of the repeater, because the feedback channel is assumed not to change. The repeater s delay d in (2) can be easily measured with a network analyzer or similar delay measurement tools [2]. 5. Simulation Results To confirm the validity of the proposed ICAN scheme, we show by simulations that the stability and signal quality of a radio repeater are affected by factors such as the repeater s gain and delay, the propagation delay on feedback paths, feedback channel characteristics, and the capability of the feedback channel estimation algorithm. For the simulation, a signal generator that produces a wideband code division multiple access (WCDMA) signal with one frequency assignment (FA) as well as a receiver that measures the EVM of the repeater s output signal were implemented on a computer [8]. P = 1,280 was used in (15). The ADC sampling rate was 50 MHz. Two types of feedback channel models were used in this work. The first feedback channel model is comprised of four multipaths with relative time delays (power levels) of 0 μs(0db), 0.02μs( 3 db), 0.08μs( 10dB), and 0.12μs( 15 db), respectively. The second feedback channel model has five multipaths with relative time delays (power levels) of 0 μs (0 db), 0.06 μs ( 5 db), 0.1μs ( 10dB), 0.2μs ( 15 db), and 0.3 μs ( 20 db), respectively. The reason to use these channel models with small delay spreads is as follows: Contrary to the BS and MS antennas facing each other for better signal reception, the transmit and receive antennas of a radio repeater should be installed not to face each other for large isolation between those two antennas. As a result, multipaths contributing to feedback oscillation of a radio repeater are formed from multiple reflections from at least two separate and remote reflectors. In addition, the angle of arrival of feedback signals to the repeater s receive antenna is narrow, because the transmit and receive antennas of a radio repeater normally have some directivity to secure large isolation contrary to the mobile station s omni antenna. Therefore, unlike a large delay spread of conventional multipaths between the BS and MS that affect the receiver s data decoding, a delay spread of surviving feedback signals contributing to feedback oscillation of a radio repeater would be small. The amplitude of the output signal of a radio repeater without a FICANS, which is normalized to the repeater s equivalent gain G, is plotted in Fig. 4 as a function of G and k 0 under the first channel model. When the gain G = 10 db, which is equivalent to the condition that G < 1, the repeater does not oscillate, as expected. With gain G = 0dB, the repeater shows some sign of oscillation but still does not oscillate. However, with gain G = 5 db and 10 db, the repeater goes into FOSC within a short period of time regardless of delay. Furthermore, Fig. 4 shows that the larger the gain G is, the faster the repeater goes into FOSC. When a limiter is used to avoid FOSC of the repeater at the initial stage, the output signal does not increase above a given level, as expected. In this case, a threshold of x(n) /G = 10 3 was used for limiting and the error signal e(n)(= x(n) /G), which was larger than the threshold, was scaled down to make a moving average of the error signal e(n) over a given window be

7 1784 IEICE TRANS. COMMUN., VOL.E92 B, NO.5 MAY 2009 Fig. 4 The amplitude of the normalized output signal x(n)/g of a radio repeater without a FICANS, as a function of G and k 0 under the first channel. Fig. 5 The amplitude of the normalized output signal x(n)/g of a radio repeater with the proposed ICAN scheme, as a function of N, G, andl 0 under the first channel. less than the threshold. As the total delay (k 0 + d) decreases, the rate of increase of the output signal rises, because the feedback signal circulates the FCL faster. The amplitude of the normalized output signal of a radio repeater with the proposed ICAN scheme is shown in Fig. 5 as a function of G, the estimated delay l 0,andthe LMS filter length N under the first channel model. Here, it was assumed that the total delay d t was 5.8 μs. The LMS algorithm with the control parameter μ = 100 was used to update the coefficients of the estimated channel filter (LMS filter). The value of μ was chosen under the condition that the input SNR was 25 db with the input signal power being 70 dbm. When G = 10 db, N = 10, and l 0 = k 0, the repeater does not oscillate, because the LMS algorithm cancels all the multipath feedback signals in the feedback channel model. However, there is some increase of the output signal at the initial period of time due to the slow convergence rate of the LMS algorithm. When G = 10 db, N = 3, and l 0 = k 0,even if the 3rd and 4th feedback signals are not directly cancelled Fig. 6 EVM performance of a radio repeater with the proposed ICAN scheme under various conditions. by the LMS filter with a length N = 3, they are partially cancelled by the three coefficients of the LMS filter. As a result, there is no FOSC of the repeater by these uncovered feedback signals due to their power levels being small, but the normalized amplitude of the repeater s output signal is shown to converge to a large value. In the case where G = 5dB,N = 10, and l 0 = k 0 + 2, even though l 0 is not equal to k 0, the LMS filter with a length N = 10 covers all feedback signals in the feedback channel model. Hence, the normalized amplitude of the repeater s output signal converges to a small value. When G = 5dB, N = 10, and l 0 = k 0 2, the repeater oscillates, because the two dominant feedback signals are not cancelled by the LMS filter at all. To check the signal quality of a radio repeater based on the proposed ICAN scheme, the noise effect on the EVM in (15) is first measured. The signal generator is connected to the input of a radio repeater that has G 0 of 10 db, input SNR of 25 db, no FICANS, and no feedback channel. The mean EVM is about 4.5%, as shown in Fig. 6. This EVM can be considered optimum from the viewpoint of a repeater with the FICANS when G > 1. When the LMS algorithm, at N = 10, μ = 100, and l 0 = k 0, is used to cancel the feedback signal under the first channel (MP1), the mean EVM of the radio repeater with G = 5dB and d t = 5.8 μs is about 4.7%. The ICAN scheme under the same conditions shows worse EVM (about 5.2%) when G = 10 db compared to the case of G = 5dB. The reason for this is that the LMS algorithm with a slow convergence rate does not closely follow the rapid increase of the output signal, where the rapid increase occurs due to the larger G. Under the same conditions and G = 10 db, the EVMs are almost the same for the two cases of d t = 5.8 μs and d t = 4 μs, because the LMS algorithm with given N and μ can cancel the feedback signals sufficiently well in both cases. Under the second channel (MP2), which has a delay spread corresponding to 16 filter taps, the ICAN scheme shows better EVM (about 5.8%) when N = 20 compared to the case of N = 15. The reason for this is the ICAN scheme with N = 15 does not cover the last feedback path. The ICAN scheme under the worse channel (MP2) shows

8 LEE et al.: AN INTERFERENCE CANCELLATION SCHEME FOR MOBILE COMMUNICATION RADIO REPEATERS 1785 worse EVM compared to the cases under the better channel (MP1). The ICAN scheme using the RLS algorithm with δ of and λ of 1 shows the best EVM (about 4.5%) owing to a faster convergence rate. It was found by simulations that the signal quality of a radio repeater based on the proposed ICAN scheme gets deteriorated when the iterative algorithm has a slower convergence rate, the repeater s gain is larger, the feedback channel condition is worse, and the length of the adaptive filter is not long enough to cover the delay spread of the feedback channel. 6. Conclusion We proposed a feedback ICAN scheme for mobile communication radio repeaters. To prevent feedback oscillation of a radio repeater with a gain that is larger than the isolation between the transmit and receive antennas, we first derived an ICAN model. Based on the ICAN model, it was shown that the stability condition of the repeater does not guarantee the quality of the repeater s output signal. We then proposed an ICAN scheme based on an iterative algorithm to meet the desired stability and signal quality of the repeater. By simulations, we confirmed that the stability and the signal quality of the radio repeater depend on the repeater s gain and delay, the propagation delay on feedback paths, feedback channel characteristics, and the feedback channel estimation algorithm. Using the proposed ICAN scheme, we obtained a mean EVM of about 6.3% for the repeater s output signal. Acknowledgments This work was supported in part by the IT R&D program of MKE/IITA under Contract No S , Korea. References [1] W.T. Slingsby and J.P. McGeehan, A high-gain cell enhancer, IEEE 42nd Vehicular Technology Conference, pp , [2] W.T. Slingsby and J.P. McCeehan, Antenna isolation measurements for on-frequency radio repeaters, Antennas and Propagation, 9th International Conference, vol.1, pp , [3] M.R. Bavafa and H.H. Xia, Repeaters for CDMA systems, IEEE Vehicular Technology Conf., pp , [4] S.J. Kim, J.Y. Lee, J.C. Lee, J.H. Kim, B. Lee, and N.Y. Kim, Adaptive feedback interference cancellation system (AF-ICS), IEEE Microwave Symposium (MTT-S) Digest, pp , [5] J. Lee, S. Park, H. Choi, Y. Jeong, and J. Yun, A design of cochannel feedback interference cancellation system using analog control, Microwave Conference, 36th European, Sept [6] D. Choi and H. Yun, Interference cancellation repeater, PCT Patent WO 2007/ A1, July [7] M. Lee, B. Keum, D.H. Woo, and H.S. Lee, A complexity reduction scheme for interference cancellation radio repeaters, 2008 International Conference on Advanced Technologies for Communication, pp , Oct [8] M. Lee, B. Keum, Y. Son, H.S. Lee, J.T. Song, and J.-W. Kim, An efficient hardware simulator for the design of a WCDMA interference cancellation repeater, 2008 IEEE VTC 68th Vehicular Tech. Conf., pp.1 6, Sept [9] M. Lee, B. Keum, J. Kim, and H.S. Lee, A radio repeater interference cancellation model for mobile communication systems, Fourth International Conference on Wireless and Mobile Communication, July [10] 3GPP. TS V5.8.0, UTRA repeater radio transmission and reception, ETSI [11] A.V. Oppenheim and R.W. Schafer, Discrete-Time Signal Processing, 2nd ed., p.29, Prentice Hall, [12] R.A. Shafik, M.S. Rahman, and A.R. Islam, On the extended relationships among EVM, BER, and SNR as performance metrics, 4th International Conf. on Electrical and Computer Engineering (ICECE) 2006, Dec [13] M.H. Hayes, Statistical Digital Signal Processing and Modeling, pp , John Wiley & Sons, [14] J.G. Proakis, Digital Communications, 4th ed., pp , Mc- Graw Hill, New York, USA, [15] S. Haykin, Adaptive Filter Theory, 4th ed., pp , , Prentice-Hall, Moohong Lee received the B.S. degree in Electronics Engineering from Hanyang University, Seoul, Korea, in 1989 and the M.S. degree in EE Engineering from KAIST, Seoul, Korea, in He worked as a RF engineer for more than 10 years. He now works as a team leader for software modem development in the MMPC, KAIST and is currently pursuing the Ph.D. degree in EECS from KAIST. His research interests include RF systems, wireless communication modems and systems. Byungjik Keum received the B.S. degree in EE Engineering from Chungang University, Seoul, Korea, in 2005 and the M.S. degree in EE Engineering from KAIST, Daejeon, Korea, in He now works for software modem development in the MMPC, KAIST and is currently pursuing the Ph.D. degree in EECS from KAIST. His research interests include digital signal processing and mobile TV, and wireless communications. Young Serk Shim received the B.S. in Electronics Engineering from Seoul National University, Seoul, in 1976 and his M.S.E. and Ph.D. in Electrical Engineering from KAIST, Korea, in 1978 and 1982, respectively. He is currently a director of the MMPC, KAIST. His research interests include image signal processing, wireless communication systems. Hwang Soo Lee received the B.S. in Electronics Engineering from Seoul National University, Seoul, Korea, in 1975 and his M.S.E. and Ph.D. in Electrical Engineering from KAIST, Seoul, Korea, in 1978 and 1983, respectively. He is currently a professor in the Department of EECS, KAIST. His research interests include signal processing, digital communications, and next-generation convergence networks.