Covariance Matrix Adjustment for Interference Cancellation Improvement in Adaptive Beamforming

SUKHONTHAPHONG et al.: COVARIANCE MATRIX ADJUSTMENT FOR INTERFERENCE CANCELLATION IMPROVEMENT 27 Covariance Matrix Adjustment for Interference Cancellation Improvement in Adaptive Beamforming Thanakorn Sukhonthaphong, Phaisan Ngamjanyaporn, student member Chuwong Phongcharoenpanich, member, and Monai Krairiksh, member Faculty of Engineering and Research Center for Communications and Information Technology, King Mongkut s Institute of Technology Ladkrabang, Bangkok 10520, Thailand Email: s4061703@kmitl.ac.th ABSTRACT This paper proposes the interference cancellation improvement of smart antenna system by using covariance matrix adjustment. This technique includes the specific adjustable multipliers in both desired signal and interference signal covariance matrices of complex weight in order to overcome some disadvantages and improve the interference cancellation efficiency in the beamforming of adaptive array. The proposed beamforming technique in this paper is based on the complex weight which uses the covariance matrix for the computation. The Applebaum array and the Linearly Constrained Minimum-Variance (LCMV) method are used in this paper. The simulation and experimental results demonstrate that the proposed technique can improve and increase the interference cancellation of smart antenna to be more efficient than the conventional technique. Keywords: Adaptive array, LCMV, Applebaum array, Beamforming, Covariance matrix, Interferencecancellation 1. INTRODUCTION Smart antennas have recently received increasing a role for using to improve the performance of wireless communication systems [1]. These antenna systems are composed of many components. The one important component is a beamforming system that attempts to enhance the desired signal and suppress the interference signals. Among various beamforming schemes of smart antenna systems, the Least Mean Square (LMS) array, the Applebaum array algorithms proposed by Widrow, et al. and Applebaum [2], respectively and the LCMV [3] method, have been attracted considerable attention by a large number of researchers. The LMS array uses the comparison between the output and the reference signals to pursue the minimization of the Mean Square Error (MSE) whereas the Applebaum array endeavors to seek the maximization of the desired signal-to-interferenceplus-thermal noise ratio (SINR). On the other hand, LCMV method try to minimize the output power of the CM5R16: Manuscript received on March 7, 2003; revised on July 24, 2003. adaptive array under constrain. The LMS array requires the desired signal waveform, however, it does not need incident angle knowledge. On the contrary, Applebaum array and LCMV can be used when the incident angle of the desired signal is known [2]-[6]. In practice, the convergent rate and the performance of the system in adaptive array are important to perform its usefulness. The Applebaum array has the advantages in its simple hardware structure and fast convergent time. However, the convergent time can be extremely long when multiple interference signals impinge on the array, causing an eigenvalue spread. Nevertheless, this disadvantage can be overcome by using the Gram- Schmidt preprocessor, the Sample Matrix Inversion (SMI) [4] and etc. These methods require large number of computations and very complex hardware. If the interference signals could be distinguished according to their relative power level, the Applebaum array and the LCMV can effectively remove all interference signals regardless of the eigenvalue spread of the input signal covariance matrix [4]. However, when the received signal power level decreases and fluctuates, the null response in the interference direction of beamforming system will be disturbed and throb. Hence, the null response direction is not exactly in the interference direction, resulting in the performance degrading on interference cancellation. According to the aforementioned interference cancellation disadvantage, this paper proposes the covariance matrix adjustment technique to solve that problem. The related work that uses the covariance matrix to improve the performance of the system was shown in [7]. However, the methodologies of these two techniques are different. The work in [7] uses the spatial covariance matrix transformation, while the proposed technique in this paper includes the specific adjustable multipliers in both desired signal and interference signal covariance matrices of complex weight. This is to increase the low power interference signals correlation and decrease the correlation between the high powers of desired signal and interfering signals. The proposed beamforming technique is based on the complex weight which uses the covariance matrix for the computation. Thus, the Applebaum array and the LCMV are used in this proposed technique. This will set exact deep null response in the interference signal directions and high response in the desired signal direction which can improve the weak signal problem.

28 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS, VOL. 1, NO. 1, AUGUST 2003 This paper includes the conventional technique of the Applebaum array and the LCMV method in section 2, covariance matrix adjustment technique in section 3, computer simulation results in section 4, experimentation in section 5 and finally the conclusions in section 6. 2. CONVENTIONAL TECHNIQUE 2.1 Applebaum Array The concept of the Applebaum array begins with the optimization criterion by considering the analytic signals xi () t of an N-element adaptive array and complex weights w, i = 1, 2,..., N as shown in Fig.1, where θ is the angle of arrival. i X = [ x ( t), x ( t),..., x ( t)] T (1) 1 2 From Fig.1, the analytic signal xi () t [8], which may consist of desired signal, interference and noise as X = Xd + Xi + Xn. (2) They are multiplied by complex weights summed to be the output signal expressed as N w i, then s () t [2], [9]. It can be T s () t = W X = s () t + s () t + s () t (3) d i n with a vector form that under narrow-band uncorrelated jamming sources assumption [2], [4]. The weight vector optimization in the Applebaum array is based on maximization of SINR where P P d d SINR = = P u P i + P n. (4) P d, P u, P i and P n are desired signal power, undesired signal power, interference signal power and noise power, respectively. 1 Pq = Ε sq() t 2 2, (5) where q can be either d, u, i or n. At the steady state, optimal weight vector of the Applebaum array converges to Wiener-Hopf equation [4] that is given as 1 * W = µ Φ U (6) opt d where µ, U d and Φ represent an arbitrary constant, desired signal and input covariance matrix, respectively. Φ can be defined as * * * * = ( X X T )= ( X T ) ( T ) ( T dxd Xi Xi XnXn ) Φ Ε Ε +Ε +Ε x () N t w N = Φ d +Φ i +Φ n. (7)... 2.2 LCMV Method x () t 3 Σ w 3 s () t x2( t) w 2 θ Fig.1: N-element Adaptive Array x1( t) This approach is to minimize the mean square output 2 ( E S ), subject to the following linear constraints: if the input X is a column of a constraint matrix C, then the output must equal the corresponding element of a specified gain vector f, the optimum weight [3] can be shown as 1 T 1 Wopt =Φ C( C Φ C) 1 f w 1. (8) From equation (6) and (8), the complex weights are composed of the covariance matrix both the Applebaum array and the LCMV methods by which their interference signal power level is important for interference cancellation beamforming. When the received Signal-to- Noise-Ratio (SNR) level is diminished [4], the output SINR pattern response of both techniques at the interference direction will be affected. The output null pattern response will not exactly be at the interference direction. In this case, the interference cancellation efficiency is decreased. Although, there are an arbitrary constant µ, which can adjust weight vector in (6), of the Applebaum array and gain vector f in (8) of the LCMV method, they can not specify the multiplier in each covariance matrix for both the interference signal covariance matrix Φ and the desired signal covariance i matrix Φ d, which have different Interference to Noise Ratio (INR) and SNR, respectively. Thus, adjusting arbitrary constant or gain vector is not the efficient solution. For this reason, when the received SNR level is small, the interference cancellation capability of the conventional technique is decreased. 3. COVARIANCE MATRIX ADJUSTMENT TECHNIQUE Since the disadvantage of the conventional

SUKHONTHAPHONG et al.: COVARIANCE MATRIX ADJUSTMENT FOR INTERFERENCE CANCELLATION IMPROVEMENT 29 techniques occurs when the SNR level of the received signal decreases or the noise increases, we propose the covariance matrix adjustment technique to improve that mentioned disadvantage. The proposed covariance matrix adjustment technique is the process that adds the adjustable multipliers to both interference signal covariance matrix and desired signal plus noise Φ i covariance matrix Φ dn, in the complex weight analysis, for controlling output null response in the interference signal directions and output peak response in the desired signal direction. However, there is the research that relates to this technique, in which the spatial covariance matrix transformation is used in the weight computation [7]. The technique is similar to the proposed technique presented in this paper, but the methodologies of both methods are different. The spatial covariance matrix transformation technique uses the inverse rectangular window to approximate the interference covariance matrix [7] which differs from the proposed covariance matrix adjustment technique that uses the adjustable multiplications of the interference covariance matrix by considering the correlation between the received signals to enforce the beam pattern of the receiving array antenna. In addition, the proposed covariance matrix adjustment technique in this paper uses the spatial smoothing technique in the DOA estimation that can predict the correlated signal impinging at the receiving array antenna. In contrast, the spatial covariance matrix transformation technique uses only Capon s method. In order to increase the low power interference signals correlation, the adjustable multiplier with Φ i should be high, meanwhile to decrease the correlation between the high powers of desired signal and interfering signals, the adjustable multiplier with should be low. Φ d Thus, the proposed adjusted covariance matrix ( Φ defined as adj dn i adj ) is Φ = BΦ + CΦ, (9) where B and C are the adjustable multipliers. The desired signal plus noise covariance matrix consisting of both desired signal term and noise term ( Φ dn =Φ d +Φn ), is multiplied by B. Meanwhile the interference covariance matrix is multiplied by C when there is the interference. Therefore B has to be low value while C has to be high value which the value of B and C depend on the value of input SNR and input INR, respectively. If there are the interference signals from many directions, the interference covariance matrix will consist of many interference covariance matrices. In this case, the directions of the desired signal and the interference signals are derived from the Direction Of Arrival (DOA) estimations [10] (Capon s minimum-variance method, MUSIC and spatial smoothing technique [11]-[14]). For example, if there are the interference from three directions, will consist of Φ, and. Hence, Φi i1 C C1 C2 C3 Φi2 Φ i 3 will consist of, and that can be expressed as Φ = BΦ + CΦ + C Φ + C Φ. (10) adj dn 1 i1 2 i2 3 i3 In this context, if each input INR of interference signal is different, each C i of interference covariance matrix will not be identical. If each of the SNR and INR increases to be larger than the upper threshold, the values of B and C will be decreased or inversely proportional to their SNR and INR values which relate to the previous revolutionary sample, respectively. On the contrary, if each of the SNR and INR decrease to be less than the lower threshold, the value of B and C will be increased or inversely proportional to their SNR and INR, respectively. Thus and Adref B = Bref (11) Ad Airef C = Cref. (12) Ai n Otherwise, if each of the SNR and INR are between the upper and the lower thresholds, B and C will change directly proportional to their SNR and INR values which relates to the previous revolutionary sample. and ref Ad B = Bref (13) Ad C Ai ref n = Cref. (14) Airef B and are the reference values of B and C, C ref respectively, which are set to be the constant values. Ad ref and Ai ref are the reference values of the desired signal and interfering signal amplitudes which can be provided by the DOA estimation at the reference situation. Ad and Ai n are the amplitudes of the received desired signal and the n th interfering signal, respectively. They can be achieved by DOA estimation instantaneously in the adaptive process. The values of B and C are considered from the input signal of the system. Thus, it is necessary to have the control operation for comparing the input signal level and its threshold. The values of B and C can be considered from Fig.2 and Fig.3 which are the examples of relation between the level of output SINR of the Applebaum array and B for various C values in the 3, 4 SNR desired signal and, 3 INR interference signal case. Fig.2 and Fig.3 illustrate that the output SINR in the desired signal direction and the output SINR in the interference direction are proportional and inversely proportional to the value of B and C, respectively. To minimize output SINR in the interference direction, and maximize the output SINR in the desired direction simultaneously, B

30 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS, VOL. 1, NO. 1, AUGUST 2003 Output SINR(dB) in Desired Direction 45.4774038 45.4774036 45.4774034 45.4774032 45.4774030 45.4774028 45.4774026 C=1 C=5 C=10 C=30 C=50 of interference signals is more than the degree of freedom, it is necessary to choose the suitable interference from the consecutive high INR level. After the adjusted covariance matrix is obtained, the complex weight can be computed. We can then see that the weight solution of the improved covariance matrix adjustment technique for the Applebaum array and the LCMV method will multiply with the input signal of each array element and are summed to be an output signal. 45.4774024 0.001 0.0015 0.002 0.0025 0.003 0.0035 0.004 B Fig.2: Comparison of Output SINR in Desired Direction versus B for Various C Values of 3 4 SNR Desired Signal and 3 INR Interference Signal Output SINR(dB) in Interference Direction -70-90 -110-120 -130 C=1 C=5 C=10 C=30 C=50-140 0.001 0.0015 0.002 0.0025 0.003 0.0035 0.004 B Fig.3: Comparison of Output SINR in Interference Direction versus B for Various C Values of 3 4 SNR Desired Signal and 3 INR Interference Signal and C can be set to 1/500 and 30, respectively, while SNR value of the desired signal and INR value of the interference signal are between 1 and 5. These B, C, SNR and INR can be set as the reference of the system. The operation of the covariance matrix adjustment technique begins with DOA receives the input signal to estimate the signal directions. The next process is the decision part for considering the number of interference signals. If there is no interference, the adjusted covariance matrix will be set as desired signal covariance matrix, then its determinant is checked. The covariance matrix 1 will be inverted ( Φ ) in the complex weight computation [2], thus the covariance matrix should be the nonsingular matrix ( de t( Φ) 0 ). The important cause of singular matrix problem is the signal reduction or the weakness of the signal [4] that can be improved by readjusting the covariance matrix to increase the values of B and C. In addition, this technique has to include the degree of freedom decision [2] to divide and choose the interference in the suitable direction, because the number of null response pattern directions that can be set for interference cancellation of the linear array system is equal to the degree of freedom (N-1), where N is the number of the array elements. However, when the number 4. COMPUTER SIMULATION RESULTS In this section, the interference cancellation performance of the covariance matrix adjustment technique applied for Applebaum array and the LCMV method is clarified by simulation results of SINR pattern response. It is found that null response can be improved in the interference signal directions. To compare the performance of the proposed covariance matrix adjustment technique with the conventional technique, the Applebaum array consisting of four isotropic elements with half wavelength apart between elements is considered. SINR Response Pattern (db) 20 0-20 -40-60 Proposed Technique Conventional Technique -90-70 -60-50 -40-30 -20-10 0 10 20 30 40 50 60 70 80 90 Angle of Arrival (deg) Fig.4: SINR Response Pattern of Four-Element Applebaum Array with -1, 4 SNR, Desired Signal with Three Interference Signals at -7, -5 and 5 with the Same 3 INR SINR Response Pattern (db) 20 0-20 -40-60 Proposed Technique:condition1 Proposed Technique:condition2 Conventional Technique:condition1 Conventional Technique:condition2-90 -70-60 -50-40 -30-20 -10 0 10 20 30 40 50 60 70 80 90 Angle of Arrival (deg) Fig.5: Comparison of SINR Response Pattern of Four- Element Array between the Conventional Applebaum Array and the Proposed Technique for Two Conditions

SUKHONTHAPHONG et al.: COVARIANCE MATRIX ADJUSTMENT FOR INTERFERENCE CANCELLATION IMPROVEMENT 31 The maximum number of incident interference signals that can be rejected is equal to the degree of freedom that is N-1 or three incident interference signals. When there are three interference signals with same INR of 3 from incident angles of -7, -5 and 5 with 4 SNR desired signal from the incident angle of -1, the SINR response pattern of the conventional Applebaum array and the proposed technique can be presented in Fig.4. It represents more effective null response setting in the interference signal directions of the proposed technique than the conventional Applebaum array. In the case that the received signal power decreases, the response pattern of beamforming in adaptive array will be affected as shown in Fig.5. Condition 1, the desired signal is in -4 incident angle with 4 SNR and three interference signals are in -7, 2 and with 1, 1 and 15 db INRs, respectively. The output SINR response pattern in the interference directions of the conventional Applebaum array fluctuates and is not accurate, when the weak signal is experienced in the conventional Applebaum array. It is found that the SINR response pattern of the proposed covariance matrix adjustment technique can still set exact null response pattern in the interference directions. In case of condition 2, the desired signal comes from 2 with 4 SNR and three interference signals come from -4, 45 o and 7 with 1, 1, and 15 db INRs, respectively. The results of the proposed interference cancellation technique are still more effective than the conventional Applebaum array, that can be clarified. In addition, the number of the array elements is an important factor that has a role to increase the interference cancellation efficiency. Since the number of array elements is increased, the degree of freedom in this Applebaum array increases in the same trend. This should increase interference cancellation efficiency because it can reject more incident interference signals that come from many directions than less array element number. It can be shown in Fig.6, which demonstrates that although the number of array elements is increased the proposed Applebaum covariance matrix adjustment technique can still set null response pattern in the interference directions better than the conventional Applebaum array. In Fig.6, condition 1, desired signal comes from -4 with 4 SNR and four interference signals come from -7, 2, 4 and with 1, 1, 15 db and 15 db INRs, respectively. Condition 2, 4 SNR of desired signal comes from 2 and four interference signals come from -, -4, 45 o and 7 with 1, 1, 15 db and 15 db INRs, respectively. By using eight-element array antenna with half wavelength array element spacing, the simulation results of the Applebaum array and the LCMV method can be shown in Fig.7 and Fig.8, respectively. In this case, the response pattern of the conventional technique can not set null response pattern in all interference directions for both condition 1 and condition 2 as defined in Fig.5, while the proposed technique can set null response pattern exactly SINR Response Pattern (db) 20 0-20 -40-60 Proposed Technique:condition1 Proposed Technique:condition2 Conventional Technique:condition1 Conventional Technique:condition2-90 -70-60 -50-40 -30-20 -10 0 10 20 30 40 50 60 70 80 90 Angle of Arrival (deg) Fig.6: Comparison of SINR Response Pattern of Five- Element Array between the Conventional Applebaum Array and the Proposed Technique for Two Conditions SINR Response Pattern (db) 20 0-20 -40-60 Proposed Technique:condition1 Proposed Technique:condition2 Conventional Technique:condition1 Conventional Technique:condition2-90 -70-60 -50-40 -30-20 -10 0 10 20 30 40 50 60 70 80 90 Angle of Arrival (deg) Fig.7: Comparison of SINR Response Pattern of Eight- Element Array between the Conventional Applebaum Array and the Proposed Technique for the Same Conditions of Fig.5 SINR Response Pattern (db) 20 0-20 -40-60 Proposed Technique:condition1 Proposed Technique:condition2 Conventional Technique:condition1 Conventional Technique:condition2-90 -70-60 -50-40 -30-20 -10 0 10 20 30 40 50 60 70 80 90 Angle of Arrival (deg) Fig.8: Comparison of SINR Response Pattern of Eight- Element Array between the Conventional LCMV Method and the Proposed Technique for the Same Conditions of Fig.5 in the interference directions. The good response pattern close to the desired direction is achieved. 5. EXPERIMENTATION 5.1 Experimental Configuration The experiments were conducted in the anechoic

32 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS, VOL. 1, NO. 1, AUGUST 2003 chamber of the Communications Research Laboratory (CRL) Japan at Yokosuka Radio Communications Research Center. The positions of desired signal and interference transmitters and receiver are depicted in Fig.9. The desired transmitting antenna is a horn and the interfering transmitting antenna is a single patch, while the received antenna is an eight-element patch antenna array with half wavelength inter-element. The π/4 QPSK modulated signal is transmitted at the carrier frequency of 2.335 GHz. The sampling rate of the receiver is 1.8 MHz and IF is 450 khz. The block diagram of the receiver can be shown in Fig.10. All eight elements of the array antenna were connected to down converters for converting the received signal in each branch of array antenna down to 450 khz IF. The IF signal in each channel passed the A/D converter for changing an analog signal to a digital signal before adaptive processed by the computer. In adaptive process, the signal from each branch was collected separately among the other branches. Receiver 10.80 m -18.46 79.13 Horn Antenna (Desired Signal) 10.70 m 3.45 m Single Patch Antenna (Interference) Fig.9: Configuration of the Experiment (not to scale) Antenna. #1.. Antenna #8 Down Down Convertor Down Convertor Convertor Down Converter LNA BPF LNA BPF LNA 2.335GHz BPF to 450kHz IF 450kHz A/D 1.8MHz Oscillator Oscillator 2.335GHz Oscillator 2.335GHz 2.335GHz Adaptive process Fig.10: Block Diagram of the Receiver Output Signal The collected data were used to estimate the direction of incident signal and the relative power intensity by using the Capon s minimum-variance method [14] before the complex weight computation. In this case, the relative power intensities in the incident directions of the received signals were used to provide covariance matrix for the complex weight computation. In order to realize the precise computing processes in both of direction estimation and complex weight computation, calibration procedure is indispensable to compensate particular amplitude and phase imbalance among RF circuits of the antenna branch [15]. Moreover, the complex weights were computed in various situations of the experiment with the same directions of the desired and the interfering signals. The experimental results are presented below. 5.2 Experimental Results In this section, the output weight in various experimental situations were computed and used to adjust the beam pattern of the receiving array antenna by multiplying the complex weight with the beam pattern of the receiving antenna for realizing the interference rejection capability. Since the Applebaum array needs the knowledge of incident angle, the direction estimation technique is required. Accordingly, to find the direction and relative power intensity of the received signal used in covariance matrix computation, the Capon s minimumvariance method is applied for the Applebaum array. At first, the radiation pattern of the receiving antenna before complex weight adjusting was measured as shown in Fig.11. In the adjusted case of two noncoherent transmitters, the beam pattern of the receiving antenna after adjusted or multiplied by complex weight of both the conventional technique and the proposed technique are compared in Fig.12. The direction of the desired signal and the interfering signal are and -18.46 o, respectively. The relative power in the desired signal and the interference directions of the conventional technique are -15.5 db and -3 db, respectively, but those of the proposed technique are -5.5 db and -15 db, respectively. In addition, in Fig.12, the direction of the desired signal is 1 and the interfering signal direction is -8.46 o. The values of the beam pattern in the desired signal direction and the interference signal direction of the conventional technique are -7 db and, respectively. On the other hand, those of the proposed technique are -7 db and -19 db, respectively. It should be noted that the beam pattern in the interfering direction of the proposed technique is deeper and more exact than the conventional technique, 12 db in Fig.12 and 9 db in Fig.12. The beam pattern in the desired direction of the proposed technique still has better response than that of the conventional technique. Moreover, the proposed technique can still be used with the LCMV method as shown in Fig.13 and, which show that the proposed technique provides the more effective in beamforming than the conventional technique. In Fig.13, the relative power in the desired signal direction and the interference

SUKHONTHAPHONG et al.: COVARIANCE MATRIX ADJUSTMENT FOR INTERFERENCE CANCELLATION IMPROVEMENT 33-3 3-3 3 - - - - 15 Fig.11: Azimuth Pattern of Receiving Antenna before Complex Weight Adjustment - -3 3 Conventional Technique -3-15 Proposed Technique 3 - - - Conventional Technique - -3 Conventional Technique 15 Proposed Technique 3 15 Proposed Technique Fig.12: Received Beam Patterns after Adaptation of Applebaum Array, Desired Signal and - 18.46 o Interference Directions, 1 Desired Signal and -8.46 o Interference Directions. Conventional Technique 15 Proposed Technique Fig.13: Received Beam Patterns after Adaptation of LCMV Desired Signal and -18.46 o Interference Directions, 1 Desired Signal and -8.46 o Interference Directions. signal direction of the conventional technique are -12.6 db and -9 db, respectively, whereas those of the proposed technique are -5.5 db and -15 db, respectively. In Fig.13, the relative power in the desired signal and the interference directions of the conventional technique are - 2.2 db and -4.5 db, respectively. On the contrary, those of the proposed technique are -6.6 db and -19.7 db, respectively. It can be concluded that the beam pattern in the interfering direction of the proposed technique is deeper and more exact than the conventional technique, 6 db in Fig.13 and 15.2 db in Fig.13. Furthermore, when the interference is coherent signal, the beam patterns of the received array antenna are presented and compared for various power levels of the interference in Fig.14. In this case, the DOA can be estimated from the spatial smoothing technique with MUSIC method [13]. The power of the desired signal was fixed while the

34 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS, VOL. 1, NO. 1, AUGUST 2003 power of the interference was varied 3 db down per one step from Exp.1 to Exp.5 by using attenuator. Therefore, the output power of interference from Exp.1 to Exp.5 are -38 db, -41 db, -44 db, -47 db and -5, respectively. The direction of the desired signal was while interfering signal was -18.46 o. Also, in Fig.15 the beam patterns of the receiving antenna were compared in the same situation of Fig.14, but the directions of the desired signal and the interfering signal were changed. In Fig. 15, the direction of the desired signal was 1 while the interfering signal direction was -8.46 o. The average powers of both conventional and proposed techniques in the desired direction of Fig.14 are -6.3 db and -3.9 db, respectively, whereas those of Fig.15 are -3.6 db and -2.9 db, respectively. In the case of interfering direction of Fig.14, the average power of both the conventional and the proposed techniques are -10. and -15.1 db, respectively, while those of Fig.15 are -7.3 db and -21.2 db, respectively. Therefore, the power level of the proposed technique in the desired direction of Fig.14 is 2.4 db higher than the conventional technique, while Fig.15 is 0.7 db higher than the conventional technique. Besides, the power level of the proposed technique in the interfering direction of Fig.14 is 5.1 db lower than the conventional technique, while Fig.15 is 13.9 db lower than the conventional technique. Thus, the descriptive results can verify the more effective of the proposed technique that the beamforming can set null beam at the interference direction and more effective response pattern - -3 3 - -3 3 - - 15-3 3 - - 15-3 3 - - 15 Fig.14: Applebaum Array Beam Pattern of Receiving Antenna for Various Interference Power Levels when Desired and Interference Directions are and -18.46 o, respectively, Conventional Technique, Proposed Technique. 15 Fig.15: Applebaum Array Beam Pattern of Receiving Antenna for Various Interference Power Levels when Desired and Interference Directions are 1 and -8.46 o, respectively, Conventional Technique, Proposed Technique.

SUKHONTHAPHONG et al.: COVARIANCE MATRIX ADJUSTMENT FOR INTERFERENCE CANCELLATION IMPROVEMENT 35-3 3-3 3 - - - - 15 15-3 3-3 3 - - - - 15 15 Fig.16: LCMV Beam Pattern of Receiving Antenna for Various Interference Power Levels when Desired and Interference Directions are and -18.46 o, respectively, Conventional Technique, Proposed Technique. in the desired direction than that of the conventional technique for various values of power level of the coherent interference. Although the directions of the desired signal and the interfering signal were changed, Fig.15 illustrates that the proposed technique can still be more effective than the conventional technique. In case of both Fig.14 and Fig.15, the values of the adjustable multipliers (B and C) were changed in the opposite manner with respect to the power of the received signals that were estimated from the Capon s minimumvariance method and spatial smoothing MUSIC method. Otherwise, the weak signal problem would happen that were experienced in the conventional technique, resulting in inaccurate null direction for interference suppression and deteriorate response in the desired direction. Fig.17: LCMV Beam Pattern of Receiving Antenna for Various Interference Power Levels when Desired and Interference Directions are 1 and -8.46 o, respectively, Conventional Technique, Proposed Technique. Furthermore, Fig.16 and Fig.17 are illustrated in the case of the LCMV beamforming with the coherent received signals. These experimental results still give the same trend as Fig.14 and Fig.15 of the Applebaum array. Those clarify that the proposed technique can be used with the LCMV method as well, and the performance is greater than the conventional technique in various situations of coherent received signal power level. 6. CONCLUSIONS This paper proposes the covariance matrix adjustment technique to improve the beamforming system of an adaptive array to solve the weak signal problem when the SNR of the received signal is low. The feature of this method is the efficiency of setting the beam peak in the desired signal and null in the interference

36 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS, VOL. 1, NO. 1, AUGUST 2003 directions. The computer simulation results and the experimental results show that the proposed covariance matrix adjustment technique can improve interference cancellation in the adaptive array beamforming method which uses covariance matrix for the complex weight computation: Applebaum array and LCMV method. Although, there are many incident interference signals as much as the degree of freedom of the system, the proposed covariance matrix adjustment technique can solve the weak signal problem for the interference rejection. Moreover, this technique can be effectively used with many element numbers of receiving array antenna and also applied with both noncoherent and coherent received signals. 7. ACKNOWLEDGEMENT The authors are grateful to Dr. Hiroyuki Tsuji and Dr. Ryu Miura of Yokosuka Radio Communications Research Center, Communications Research Laboratory (CRL), Japan, for their invaluable comment. The Public Management, Ministry Home Affairs, Posts and Telecommunications (MPHPT) is acknowledged for supporting this research. REFERENCES [1] M. Chryssomallis, Smart Antenna, IEEE Antennas and Propagation Magazine, Vol. 42, No. 3, pp. 129-136, 2000. [2] R. T. Compton, Jr., Adaptive Antennas Concepts and Performance, Prentice-Hall, Englewood Cliffs, NJ, 1988. [3] B. R. Breed, A Short Proof of the Equivalence of LCMV and GSC Beamforming, IEEE Signal Processing Letters, Vol. 9, No. 6, pp. 168-169, 2002. [4] K. M. Lee and D. S. Han, AGC Applebaum Array for Rejection of Eigenvalue Spread Interferences, IEICE Transactions on Communications, Vol. E84- B, No. 6, pp. 1674-1678, 2001. [5] Y. Ogawa and T. Ohgane, Advances in Adaptive Antenna Technologies in Japan, IEICE Transactions on Communications, Vol. E84-B, No. 7, pp. 1704-1712, 2001. [6] G. V. Tsoulos, Smart Antennas for Mobile Communication Systems: Benefits and Challenges, Electronics & Communication Engineering Journal, Vol. 11, No. 2, pp. 84-94, 1999. [7] K. Hugl, J. Laurila and E. Bonek, Downlink Beamforming for Frequency Division Duplex Systems, Proceedings of the IEEE Global Telecommunications Conference (GLOBECOM 99), Vol. 4, pp. 2097-2101, 1999. [8] H. Krim and M. Viberg, Two Decades of Array Signal Processing Research, IEEE Signal Processing Magazine, Vol. 13, No. 4, pp. 67-94, 1996. [9] S. P. Applebaum, Adaptive Arrays, IEEE Transactions on Antennas and Propagation, Vol. AP-24, No. 5, pp. 585-598, 1976. [10] B. Friedlander and B. Porat, Performance Analysis of a Null-Steering Algorithm Based on Direction-of- Arrival Estimation, IEEE Transactions on Acoustics, Speech, and Signal Processing, Vol. 37, No. 4, pp. 461-466, 1989. [11] J. S. Jeong, K. Sakaguchi, K. Araki and J. Takada, Generalization of MUSIC Using Extended Array Mode Vector for Joint Estimation of Instantaneous DOA and Angular Spread, IEICE Transactions on Communications, Vol. E84-B, No. 7, pp. 1781-1789, 2001. [12] J. Choi, I. Song, S. I. Park and J. S. Yun Direction of Arrival Estimation with Unknown Number of Signal Sources, Proceedings of the ICCS/ISITA 92, pp. 30-33, 1992. [13] B. Friedlander and A. J. Weiss, Direction Finding Using Spatial Smoothing With Interpolated Arrays, IEEE Transactions on Aerospace and Electronic Systems, Vol. 28, No. 2, pp. 574-587, 1992. [14] W. Featherstone, H. J. Strangeways, M. A. Zatman and H. Mewes, A Novel Method to Improve the Performance of Capon's Minimum Variance Estimator, Tenth International Conference on Antennas and Propagation, Vol. 1, No. 436, pp. 322-325, 1997. [15] M. Oodo and R. Miura, A Remote Calibration for a Transmitting Array Antenna by Using Synchronous Orthogonal Codes, IEICE Transactions on Communications, Vol. E84-B, No. 7, pp. 1808-1815, 2001. Thanakorn Sukhonthaphong was born in Nakhonratchasima province, Thailand, in 1980. He received B.Eng. from School of Telecommunication Engineering, Suranaree University of Technology (SUT), Nakhonratchasima, Thailand, in 2001. He is currently pursuing the Master s degree at the Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang (KMITL), Bangkok, Thailand. In 2002, he did the research with the Wireless Innovation Systems Group of the Communication Research Laboratory (CRL), supporting by the Public Management, Ministry Home Affairs, Posts and Telecommunications (MPHPT), Japan. His research interests include smart antenna, beamforming system of adaptive array antenna and DOA estimation. Phaisan Ngamjanyaporn was born in Chonburi province, Thailand, in 1977. He received B.Eng. and M.Eng. from Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang (KMITL) in 1998 and 2001, respectively. He is currently pursuing the D.Eng. degree at the same institute. He is granted by the Thailand Research Fund (TRF) through the

SUKHONTHAPHONG et al.: COVARIANCE MATRIX ADJUSTMENT FOR INTERFERENCE CANCELLATION IMPROVEMENT 37 Royal Golden Jubilee Ph.D. program (RGJ-Ph.D.). His research interests include switched beam antenna, phased array antenna and antenna for mobile communication systems. Chuwong Phongcharoenpanich was born on Sept.11, 1974. He received B.Eng., M.Eng. and D.Eng. from Faculty of Engineering, King Mongkut s Institute of Technology Ladkrabang (KMITL) in 1996, 1998 and 2001, respectively. He is currently a lecturer at Department of Telecommunication Engineering, King Mongkut s Institute of Technology Ladkrabang (KMITL) and serves as the assistant leader of Wireless Communication Laboratory, Research center for Communications and Information Technology at the same institute. His research interests are antennas for mobile and wireless communications, conformal antennas and array theory. He is a member of ECTI, IEICE and IEEE. Monai Krairiksh was born in Bangkok. He received B.Eng., M.Eng. and D. Eng. from King Mongkut s Institute of Technology Ladkrabang (KMITL) in 1981, 1984 and 1994, respectively. In 1981, he joined the KMITL and is presently an associate professor in the Department of Telecommunication Engineering and serves as the leader of Wireless Communication Laboratory, Research Center for Communications and Information Technology at the same institute. His main research interests are in antennas for mobile communications, steerable beam antenna and microwave for biological and industrial applications. He is a member of ECTI, IEICE and IEEE.