Study of High-Accurate Frequency Estimation in 60GHz Wireless Communication System

Transcription

1 Journal of Communications Vol. 10, o. 7, July 015 Study of High-Accurate Frequency Estimation in 60GHz Wireless Communication System Kun Chen 1, Yingxin Zhao 1, Hong Wu 1, Qiqi Wang 1, Yong Liu 1, Ran Tang 1, Bing Liu 1, and Lijun Ge 1. Tianjin Key Laboratory of Photonics Materials and Technology for Information Science, College of Electronic Information and Optical Engineering, ankai University, Tianjin , China. School of Electronics and Information Engineering, Tianjin Polytechnic University, Tianjin , China {kunchen87, randytang009nkdx, liubing.job, {zhaoyx, Abstract A high-accurate frequency estimation algorithm is proposed in the 60GHz OFDM wireless communication system. The phase offset is caused by the carrier frequency offset, and its accumulation has a linear relationship in the time. So the preamble sequence phase can be obtained by the algorithm in this paper, and then the carrier frequency offset value is calculated by the linear regression theory. The 60GHz wireless communication system simulation platform is established. In the platform, the algorithm is simulated with different carrier frequency offset, and the algorithm estimation performance is contrasted with the traditional algorithm. The results show that accuracy of the algorithm has 30dB higher than the autocorrelation algorithm, and that the BER performance of the algorithm is the same as witch in the case of no frequency offset. So the high-accuracy of frequency offset estimation is proposed by the algorithm in the 60GHz wireless communication. Index Terms Linear regression, frequency offset estimation, OFDM, 60GHz wireless communication I. ITRODUCTIO Your goal is to simulate the usual appearance of papers in a Journal of the Engineering and Technology Publishing. We are requesting that you follow these guidelines as closely as possible. With the development of wireless communication technology, people have demand for higher data rate of wireless transmission [1]. When the carrier frequency of WiFi is.4/5ghz, the maximum transmission rate of WiFi is only 1.3Gbps. However the wireless transmission rate in the IEEE80.11ad-01 amendment [] will enhance to 7Gbps, and it can be used in the uncompressed HD video wireless transmission. But the carrier of IEEE80.11ad-01 amendment is used in 60GHz band, and its band penetration is poor. So in the future, to meet the high speed and penetration Manuscript received March 19, 015; revised July 0, 015. This work was supported by the Joint Specialized Research Fund for the Doctoral Program of Higher Education ( ), Tianjin city application basis and cutting-edge technology research projects(13jczdjc6000, 13JCQJC01000). The study is supported by the ational atural Science Foundation of China (o ), the atural Science Foundation of Tianjin for Young Scientist (o. 13JCQJC00900). Corresponding author zhaoyx@nankai.edu.cn doi:10.170/jcm requirements, the wireless transmission system which the carrier frequency of the system can be switched in.4, 5 and 60GHz will become the mainstream scheme of the market. In order to achieve the 7Gbps transmission rate, the OFDM modulation is used in IEEE80.11ad-01 amendment system, but the OFDM system sensitizes to carrier frequency offset [3] which is caused by the frequency deviation of the crystals between the transmitter and receiver. Auto-correlation algorithm is presented in paper [4], in which frequency offset is estimated by ug the phase of the preamble sequence auto-correlated value. By ug the auto-correlation algorithm, there is not training sequence stored in the internal memory, so this algorithm can save the hardware resources. But it is susceptible to effects of noise and multi-path, and it has less precision when it is used in the large amount of data transmission system. A frequency offset estimation algorithm of linear regression in the gle carrier UWB system was presented in paper [5]. In this algorithm, the phase is extracted by ug the 18 P continuous sequence of cross-correlation value, then the frequency offset is estimated by ug the linear relationship between each phase angle. However the algorithm only can be used in gle carrier systems. In the multi-carrier system, phase value obtained by algorithm of paper [5] has no linear relationship, thus the frequency offset can not be estimated. The algorithms which mentioned in the paper [6] and [7] can not meet the high precision of the system frequency offset estimation. So in this paper, the high precision frequency offset estimated algorithm is proposed for ug linear regression theory, which also can be used in the OFDM system. II. LIEAR REGRESSIO FREQUECY OFFSET ESTIMATIO ALGORITHM A. Frame Structure in the 60GHz System OFDM physical layer frame structure of the IEEE80.11ad-01 amendment is comprised by the leader sequence, OFDM data and the pilot. The leader sequence including the long training sequence and short training sequence, is used to achieve the synchronization and channel estimation. There are 17 Golay 474

2 Journal of Communications Vol. 10, o. 7, July 015 complementary sequence (called Ga) in the short training sequence, which the each Ga has 18 bits, and long training sequence is constituted by Gv, Gu and -Gb. Ga and Gb is a group of Golay complementary sequence pair. The frame structure and the leader sequence components is shown in Fig. 1. Gv = [-Gb Ga -Gb -Ga] Gu = [-Gb -Ga Gb -Ga] Ga Ga Ga Ga Ga Gv Gu -Gb CP Header CP Data CP Data Fig. 1. The frame structure offset ε. With the increase of n, the phase angle a[n] is periodic with π, and in each period, the phase is a linear function, as shown in Fig.. To eliminate the periodic, a[n] is necessary to be processed by data connection, as shown in Fig. 3. Through the preprocessed a[n], the normalized frequency offset is obtained by ug least squares linear regression. In the 60GHz standard, there are 338 points of preamble sequence in a frame. In order to reduce the complexity and interference of data connected algorithm, only the first 640 point is selected to calculate the normalized frequency offset ε1. B. Phase Extraction The system model is set to L 1 y[n] (h[i]s[n i]) e ( j n j ) w[n] (1) i 0 where s[n] is the received leading sequence, and s[n] is the transmitted leading sequence, h[i] is the channel impulse response, w[n] is the complex of Gauss white noise, L is the length of the channel impulse response, ε is the normalized frequency offset, is the number of subcarriers. In this system, the value of is 51. And θ is the initial phase offset. The system has completed the frame synchronization in this paper. In AWG channel, model can be simplified as (). y[n] s[n]e ( j n j ) w[n] Fig.. Receive sequence phase offset () The locally stored preamble sequence s[n] conjugate multiplies with the received preamble sequence s[n], then the phase angle a[n] which is caused by frequency offset can be obtained. a[n] angle( s[n]s* [n]) j n j angle s[n] e s[n]w* [n] (3) n [ n] If the phase (γ [n]) caused by the noise can be ignored, the frequency offset phase angle of the received sequence is the (4). a[n] n Fig. 3. After the connection of the phase In the case of low SR, the continuation algorithm of eliminating period will be disturbed by the noise, and that will lead the processed a[n] into sections, as shown in Fig. 4, therefore the linear regression calculation of the frequency offset error is too large. So it is necessary to check the piecewise before ug the linear regression calculation. The frequency offset is calculated in each segment by linear regression, and then the final frequency offset is obtained by mean procesg the each segment values. Through the theoretical calculation and experimental simulation, the linear regression of each segment of the points should be larger than Q in order to ensure the accuracy of the frequency offset estimation. And if the length of the segment is less than Q, this segment will be abandoned. (4) C. Linear Regression Frequency Offset Estimation Algorithm By the formula (4), frequency offset phase (a[n]) of receiving sequence is a linear function which n is the argument and / is the slope. If the slope of the function can be calculated, the normalized frequency offset ε would also be known. So the main work of this paper is to calculate the slope of receive sequence frequency phase angle a[n] by ug linear regression method and then calculate the normalized frequency 475

3 Journal of Communications Vol. 10, o. 7, July 015 Only the preamble sequence is frequency offset compensated by ug ε1, and then the phase of the whole preamble sequence can be obtained by the algorithm. By the formula (5), if the 338 points are used in the linear regression calculation, and the number of sub-carrier is 51, the residual normalized frequency offset would be less than var c1 n n 1 (8) According to (8) and simulation results, the linear regression calculation, each segment of data points should not be less than 18 points, so in the.3 section Q shall be taken as 18, in order to ensure the frequency deviation is less than 1ppm. III. THE IMPROVED ALG ORITHM I LOW SR AD MULTI-PATH CHAEL When SR is low, the performance of the system is greatly influenced by the noise as formula (3). So the linear regression algorithm should be improved. The correlation algorithm of point M=18 is calculated with the received sequence y[n] and transmitted sequence s[n], where M is length of the Ga sequence. Because the mode value of Golay complementary sequence is 1 that s[n+k] =1, where: Fig. 4. Successive phase of 3dB SR R[k ] (5) j k j M 1 D. The Accuracy of the Frequency Offset Estimation The linear regression equation of the phase angle can be taken as a=c0+c1k among the a as phase, k as the preamble sequences points c1 as normalized frequency offset. So c1 can be got by linear regression, as ki k ai a a[k 1] a[k ] (6) i 1 k j k ] var ai a[k ] M 1 s[n k ]w*[n k ] (9) dw[k ] (10) 1 n n 1 k a[0] dw[k ] k a[0] (11) Because initial phase as a[0] is irrelevant with k and it is the intercept of linear function in (11), the method is not sensitive to noise. So the normalized frequency offset ε can be obtained by a[k] sequence, which the method is linear regression calculation. In a relatively stable environment, multi-path channel impulse response h[i] varies slowly, and its length L is (7) j 1 So the estimation accuracy has relationship with the system noise σ and the number of participating points in the linear regression calculation. the system noise variance σ can be considered a constant. So var(c1) and n follows (8). Thus, a[k] has linear relationship with k. s as n. Because k is continuous natural numbers, k is the middle of the k. The variance of c1 is var(c1), as: ki k j n j πεk jθ s k, the number of the linear regression calculation point n The phase of cross-correlation R[k] as a[k], phase of noise as aw[k], dw[k+1]=aw[k+1]-aw[k], where the phase expectation of Gauss noise is 0, as dw[k] 0. So the phase difference of R[k+1] and R[k] as formula (10), where var(ai) is variance of gauss noise as σ, mean of k a n j πεm 1 e M 1 s[n k ]w* [n k ] j πε 1 e πεm j πε ( k M 1) M 1 jθ s[n k ]w* [n k ] πε n 0 frequency offset compensated by the ε. var (c1 ) [ s[n k ] e The residual frequency of the preamble sequences is less than the value above. So there does not need connected algorithm with the 338 points, and the residual frequency offset ε can be directly calculated by the linear regression. The total normalized frequency offset is 1. Then the hole dates can be ki k s[n k ] y*[n k ] 338 c1 M 1 L 1 limited. So h* [i] is a constant value, and its mode is i 0 Rh, and angle is ah. Similarly, the cross-correlation function as RMP[k] can be derived from formula (1). 476

4 Journal of Communications Vol. 10, o. 7, July 015 RMP [k ] M 1 calculated ug 18 points cross-correlation, and then frequency offset can be calculated by linear regression. According to the standard, the range of normalized frequency offset is [-, ], and frequency offset is estimated as shown in figure 6. Then value of frequency deviation estimation Mean Square Error (MSE) is calculated. In the Gauss channel model, the MSE values of the 1st algorithm, nd algorithm, and the AutoCorrelation (AC) algorithm in paper [4] is contrasted in Fig. 6. It is shown in Fig. 5, the two algorithms in this paper can estimate the frequency offset value in each frequency offset value, but the auto-correlation algorithm can not estimate the frequency offset value while the value is too large. It can be learned from Fig. 7 that 1st algorithm is obviously disturbed by the noise. In the nd algorithm, the extracted phase of cross-correlation function is not obvious obstructed by noise, so the linear relationship is not broken. The accuracy of the nd algorithm is still 30dB higher than the auto-correlation algorithm at each frequency offset values. Therefore, the estimated value of nd algorithm is more stable. s[n k ] y*[n k ] n 0 j k j M 1 e j n L 1 * M 1 h [i] s[n k ]w*[n k ] i 0 n 0 j n M 1 (1) e Rh e jah s[n k ]w*[n k ] πεm j πε ( k M 1) M 1 jθ jah s[n k ]w* [n k ] Rh e πε In multi-path channel, from the extracted phase of correlation function RMP we can see that (9) and (1) have the same form which means phase of RMP which calculated from (1) also satisfies a linear relationship, as shown in Fig. 5. The phase of noise and multi-path are in the intercept term of the function, so the slope calculation is not affected by it. The normalized frequency offset can be calculated by the same method as (11). So this algorithm can effectively cancel multi-path interference, and it can be used in multi-path channel frequency offset estimation. j k j M 1 Fig. 6. The estimated value MSE of the three algorithms in the Gaussian channel Fig. 5. The estimated value of the three algorithms in the Gaussian channel IV. THE RESULTS OF SIMULATIO The simulation bases on the IEEE80.11ad-01 amendment 60GHz channel. In this system, it will take the channel coding ug 1/ LDPC, 16QAM modulation, 51 point sub-carriers, the length of CP is 18bit, and the leader sequence with 338 bit, frame structure as shown in Fig. 1. The source rate is.77gbps. In the paper, the algorithm in the second section named 1st algorithm, the algorithm in the third section named nd algorithm. Fig. 7. Modulus value of the 60GHz living-room channel impulse response A. Simulation in AWG Gauss Channel Model In the 1st algorithm, firstly, frequency offset ε1 can be estimated the first 640 points with the linear regression calculation. Then the second linear regression frequency offset ε can be calculated by the compensated 338 preamble sequence. Finally, the total estimation is ε=ε1+ε. In the nd algorithm, the first 190 points is B. Simulation in 60GHz Channel Model The 60GHz channel models are described in paper [8] and [9], such as the elevation, reflection loss[10], shelter and other factors in transmission path channel. The 60GHz channel is divided into Living_room model, 477

5 Journal of Communications Vol. 10, o. 7, July 015 Cubicle model, and Conference_room model. The living room model is the most complicated in 60GHz channel model which has no direct. The channel impulse response is shown in Fig. 7. Simulation is based on the living room channel model in this paper. Compared with Fig. 4, the simulation of 60GHz channel is shown in Fig. 8. curve of the nd algorithm is below the curve of autocorrelation algorithm. This means frequency offset estimation performance of the nd algorithm is better than auto-correlation algorithm. The BER curve of nd algorithm with frequency offset has the same curve as the system without frequency offset, so nd algorithm can satisfy the frequency estimation precision of the entire system. Fig. 8. The estimated value of the three algorithms in the 60GHz living room channel Fig. 10. The contrast of frequency offset estimation algorithm bit error rate in the system V. COCLUSIO In this paper, according to the frame structure provisions of the IEEE80.11ad-01 amendment, the frequency offset can be estimated in the frame head. The preamble sequence is processed by cross-correlation and then its phase can be extracted. Finally, the normalized frequency offset is estimated by the least square linear regression algorithm. The estimation accuracy of this algorithm is 30dB higher than the commonly used correlation algorithm. The algorithm has good estimation precision in different carrier frequency offset and the multi-path fading channel. As only the frame head leader sequence is used in this algorithm, the system resources is saved. The algorithm not only can be used in 60GHz wireless communication system, but also can be widely used in other OFDM systems, in which the characteristics of high precision of frequency offset estimation will be developed. Fig. 9. The estimated value MSE of the three algorithms in the 60GHz living room channel In multi-path 60GHz channel, with the [-, ] times normalized carrier frequency offset, the frequency offset is estimated by three algorithms as shown in Fig. 9. The simulation results are shown that 1st algorithm and nd algorithm can get the correct trend of frequency offset estimation in the larger normalized frequency offset. But the auto-correlation algorithm has been unable to work. Comparing Fig. 6 and Fig. 9, the nd algorithm has the same MSE in the Gauss channel and 60GHz channel that means the nd algorithm is not sensitive to multi-path interference. Based on the simulation of two channel results, nd algorithm has a good ability of anti noise and anti multi-path. REFERECES [1] [] C. System Simulation The whole system simulation is respectively tested by ug the autocorrelation algorithm in paper [4] and the nd algorithm. There are bits useful information data sent in the transmitter, and then the Bit Error Rate (BER) is calculated in the receiver. The BER with frequency offset or not, and with Gauss channel or the living room channel model were compared in this simulation, as shown in Fig. 10. It is shown in Fig. 9 that [3] [4] 478 L. Zhuo and. Guo, 60GHz millimeter-wave wireless communications technology research, Information Technology & Standardization, no. 11, pp , ov IEEE Std 80.11ad -01, Part 11: Wireless LA Medium Access Control (MAC) and Physical Layer (PHY) Specifications Amendment 3: Enhancements for Very High Throughput in the 60 GHz Band, 01. P. C. Zhu, A carrier frequency offset estimation algorithm for OFDM systems, in Proc. Wireless Communications and Signal Procesg, IEEE, anjing, ov. 011, pp A. L. Xu, J. Wang, and K. W. Peng, Evaluation of two carrier frequency offset estimation algorithms in TDS-OFDM systems, TV Technologies, vol. 31, no. 1, pp. 7-10, Dec. 007.

6 Journal of Communications Vol. 10, o. 7, July 015 [5] S. X. Fang, D. P. Jin, L. Su, L. G. Zeng, et al., ovel frequency offset estimation algorithm based on linear regression in UWB system, Journal of Data Acquisition and Procesg, vol. 7, no. 1, pp , Jan. 01. [6] T. J. Yang and L. Hu, An improved frequency offset estimation algorithm for OFDM system, in Proc. Information etworking and Automation, IEEE, Kunming, Oct. 010, pp [7] M. M. Khairy, A novel frequency offset estimation technique for Mobile WiMAX, European Transactions on Telecommunications, vol., no. 1, pp , Jan [8] IEEE P80.11 Wireless LAs, IEEE /0334r8. Channel Models for 60 GHz WLA Systems, [9] IEEE P80.11 Wireless LAs, IEEE /0433r. PHY/MAC Complete Proposal Specification, [10] M. W. Jung, J. H. Kim, Y. K. Yoont, et al., Analysis of reflection and scattering characteristics at the 60GHz frequency, in Proc. 16th International Conference on Advanced Communication Technology, Pyeongchang, Feb. 014, pp communications. Kun Chen received the B.E. degree in communication engineering from Tianjin Polytechnic University, Tianjin, China, in 009 and the M.E. degree in communication engineering from Tianjin University, Tianjin, China, in 01. And he is pursuing his Ph.D. degree in electrical science and technology from ankai University, Tianjin, China. His research interests are in the areas of wireless Yingxin Zhao received the B. S. degree of electrical science and technology from ankai University, Tianjin, China, in 005; the M.E. degree in radio electrics from ankai University, Tianjin, China, in 009; the Ph.D. degree in electrical science and technology from ankai University, Tianjin, China, in 01. Her research direction is wireless communication technology. She is the lecturer of the ankai University. Hong Wu received the B. S. degree of electrical science and technology from ankai University, Tianjin, China, in 1989; the M.E. degree in radio electrics from ankai University, Tianjin, China, in 1995; the Ph.D. degree in control theory and control engineering from ankai University, Tianjin, China, in 005. Her research direction is wireless communication technology. From 009 to 010, she was a visiting scholar at the University of Miami in US, mainly engaged in research work of auxiliary satellite positioning. She is the professor of the ankai University. positioning. Qiqi Wang received the B. E. degree in communication engineering from ankai University, Tianjin, China, in 008; the M.E. degree in communication engineering from ankai University, Tianjin, China, in 01. And he is pursuing his Ph.D. degree in electrical science and technology from ankai interests are in the areas of auxiliary satellite Yong Liu received the B. E. degree in communication engineering from ankai University, Tianjin, China, in 013. And he is pursuing his M.E. degree in communication and information system from ankai interests are in the areas of wireless communications especially 60GHz communications System. Ran Tang received the B. E. degree in communication engineering from ankai University, Tianjin, China, in 013. And he is pursuing his Ph.D. degree in communication and information system from ankai interests are in the areas of wireless communications. Bing Liu received the B.E. degree in telecommunication from Xi an University of Posts & Telecommunications, Xi an, China, in 001 and the M.E. degree in microelectronics from anjing Electronic Devices Institute, anjing, China, in 004. And he is pursuing his Ph.D. degree in electrical science and technology from ankai interests are in the areas of wireless communications. Lijun Ge received the B. S. degree of electrical science and technology from ankai University, Tianjin, China, in 006; the Ph.D. degree in electrical science and technology from ankai University, Tianjin, China, in 011. He research direction is wireless communication technology. She is the associate professor of the Tianjin Polytechnic University. 479