Multi-GI Detector with Shortened and Leakage Correlation for the Chinese DTMB System. Fengkui Gong, Jianhua Ge and Yong Wang

788 IEEE Transactions on Consumer Electronics, Vol. 55, No. 4, NOVEMBER 9 Multi-GI Detector with Shortened and Leakage Correlation for the Chinese DTMB System Fengkui Gong, Jianhua Ge and Yong Wang Abstract As Guard Interval (GI), Pseudo Noise (PN) sequence is padded between adjacent data symbols in Chinese digital television terrestrial broadcasting (DTTB) standard. In order to accommodate with versatile applications, three GI modes are recommended with different lengths and formats. Obviously, a robust multi-gi detecting method is a prerequisite for further demodulation, such as frame synchronization, timing recovery, carrier recovery, and channel estimation. A novel multi-gi detector with shortened and leakage elation (SLC-MGI) is proposed, which adopts a modified Pre-Pos elation to overcome variable phases defined in DTMB system. Simulation results show that the proposed scheme outperforms the conventional Mixed- MMLS algorithm over both and multi-path channels, and is robust to frequency offset. Index Terms DTMB, Multi-GI detector, shortened and leakage elation, FPGA I. INTRODUCTION Digital television (DTV) broadcasting has been a very popular topic in the wireless world for years. Among many ways used for DTV services, digital television terrestrial broadcasting (DTTB) is considered to be of the most importance and challenge. As well known, in order to cope with a variety of propagation conditions encountered in the wireless broadcasting applications, a well-designed system should be customized with many optional parameters to accommodate different service requirements. In particular, for the Chinese DTTB standard - Digital terrestrial/television multimedia broadcasting (DTMB) [], there are three GI padding modes configured, which provide great feasibility for versatile service requirements. Simple and high-reliability GI detection algorithm is fairly important for such a multi-gi system as DTMB. In fact, for common DTMB receivers, it must have the knowledge of the current GI mode firstly in order to perform subsequent frame synchronization, timing recovery, and channel estimation etc. Different from the conventional technique of Cyclic-Prefix Orthogonal Frequency Division Multiplexing (CP-OFDM), the PN-aided DTMB system adopts time domain This work was supported by the National Natural Science Foundation (67738), the Natural Science Foundation of Shannxi Province (7F7), the Foundation of State Key Lab. of ISN (ISN93, ISN944) and Supported by the Project (B838). Fengkui Gong is with the State Key Laboratory of ISN, Xidian University, P.O. Box 5, Xi an 77, China (e-mail: fkgong@xidian.edu.cn). synchronization OFDM (TDS-OFDM), where GI is padded with PN sequence. Many PN-based algorithms on frame synchronization, timing recovery, and channel estimation as described in [], [3], and [4] respectively have been proposed for DTMB receivers. Although they can work well in most conditions, they all assume PN mode has been known precisely. To the best of our knowledge, only a few papers regard Multi-GI detection as a special problem for further research. In fact, analysis on GI/PN detection algorithms is worthwhile considering the requirement on low complexity and automatic acquisition of the next generation demodulation chips. Recently, a detector called as multi-mode local sequence (MMLS) detection is proposed in [5]. MMLS utilizes a new forming local elation sequence to elate with the received signal. According to our analysis, above methods using local elation have some deficiency on complexity and performance. Different from the methods using local elation, a novel simple method based on a Pre- Pos elation is proposed in this paper. In our realization, the Multi-GI detection is greatly improved by the proposed Shortened and Leakage Correlation (SLC) technique. Leakage elation means that only two modes (PN4 and PN945) are utilized to make a final decision of all three modes (PN4, PN595, and PN945), thus it can simplify the calculation by ignoring PN595 mode. Shortened elation means that only a shortened PN sequence is used for Pre-Pos elation, which can decrease the complexity and facilitates the decision. Correlation, peak search, and compare decision are three main modules of our new detector. It has been shown that only a few registers, adders and comparators are required for a reliable detection when it is realized by FPGA. By computer simulation, we will illustrate that this novel Multi-GI detection method for DTMB system is robust to noises, multi-paths and frequency offsets. The rest of this paper is organized as follows. Section II describes the frame structure of the DTMB signals and introduces the problem of Multi-GI detection. In section III the proposed Multi-GI detection method is addressed in detail. False alarm performances on various conditions are evaluated compared with typical Mixed-MMLS in Section IV. II. BASIC DESCRIPTIONS A. Signal frame structure of the DTMB system In the physical layer of the DTMB system, frame heads (FH) of PN sequences are padded between frame bodies (FB) as GIs for either multi-carrier or single-carrier block transmission. There are three major FH modes in order to Contributed Paper Manuscript received August 5, 9 98 363/9/$. 9 IEEE

F. Gong et al.: Multi-GI Detector with Shortened and Leakage Correlation for the Chinese DTMB System 789 support different service applications. The GI lengths of three modes are 4, 595 and 945 symbols. For simplicity, we call them as PN4, PN595, and PN945 respectively. PN4 and PN945 are constructed as Fig. and (c) respectively. For PN4 mode, the first L =65 symbols are same with the last 65 symbols with a fixed interval of L pn =55. PN945 mode has a similar property: the first L =434 symbols are same with the last 434 symbols with a fixed interval of L pn =5. To simple our description, we define these repeated regions with length L as copy-regions, which are combined with L pre pre-fix and L pos pos-fix symbols as illustrated in Fig.. This characteristic can be ignored in this paper for it is unhelpful for our following algorithm. ). How to combine the three GI modes into a simple algorithm, especially when PN595 is different with others? ). How to determine the optimum threshold if any? B. Mathematic Description of copy-region elation Define the received signal with symbol rate as L jπ nδfts r = h( l) x( n l) e + w () l= Where x(n), h(l), f, T s, and w(n) are the transmitted signal, the amplitude of the l-th multipath signal, the frequency offset, sampling rate, and noise, respectively. L is the number of multi-paths. The Pre-Pos elator s n-th output using copy-regions can be expressed as L * ( ) = ( + ) ( + pn + ) R n r n k r n L k k = L c( n k) () = + k = Then, the (n+)-th elation value can be calculated as L ( + ) = ( + + ) R n c n k k = L c( n k) c( n L ) c (3) k = ( ) ( ) ( ) = + + + = R n + c n+ L c n From (3), it can be seen that the Pre-Pos elator can be realized in series based on the previous elation value. (c) Fig.. Three frame heads defined in the DTMB system PN4, PN595, (c) PN945. Another characteristic related with Multi-GI detection is the FHs may be different for the different signal frames within the same super frame, serving for the identification purpose. For example, 55 PN4 sequences with different initial phases are defined, with the index from to 54. Variable phases can raise lots of difficulties as local elation is used. Instead of local elation, Pre-Pos elation utilizing copyregions is considered mainly in this paper. But for PN595 mode, since the head is constructed with a part of 3 m- sequence, there is no any copy-region if inter-frame property is not been considered. Inter-frame elation is not preferred due to the fact that it requires large hardware resources, especially too many RAM units. If elation of above copy-regions can be utilized, then a simple GI-modes detector can be realized with greatly reduced resources. All problems we need to solve can be concluded as follows. III. PROPOSED SLC-MGI ALGORITHM A. Conventional MMLS detector For methods based on local PN elation, mixed MMLS detector has advantages on both performance and complexity, just as pointed in [5]. Mixed MMLS detector use a new forming local PN sequence with length 5, which can be expressed as PNMixMMLS = [ PN4, PN595] + [ PN945,7 ] PN MixedMMLS is elated with the received baseband signals, and peak detector finds the position d n with the maximum elation value. The MMLS detector makes a decision according to the interval Δ of the adjacent d n and d n-. PN 4 if Δ< 4344 Mode = PN595 if Δ< 4495 and Δ 4344 PN945 if Δ 4495 Above decision regions are determined by considering the frame lengths of the three GI modes are different and the ranges of their phase shifts are non-overlapped. For more details, the readers can refer to [5].

79 IEEE Transactions on Consumer Electronics, Vol. 55, No. 4, NOVEMBER 9 B. Realization description of the proposed SLC-MGI For the Multi-GI detector interested, its functional diagram can be illustrated as Fig.. The received Intermediate Frequency (IF) signal is sampled by A/D and taken as the input of the Down Converter for generating the baseband I/Q signal. The Multi-GI detector utilizes the baseband signal and a fixed NCO to make a judgment. Estimated GI mode is taken as a prerequisite for synchronization. Fig.. Functional description of Multi-GI detector in the DTMB receiver. The low-complexity Multi-GI detection method based on () and (3) can be described as following. Step (Symbol Selector). Aided by the symbol-rate enable signal generated by NCO, the received symbol-rate signals r(n) are shifted to a shift-register with a length of L d =5. The NCO works as a data selector, which can be illustrated as Fig.3. This kind of symbol selector can also be used in conventional Mixed-MMLS detector. The constant W should be calculated as f W = s, fsym where f s is the sampling rate of A/D, f sym is the symbol rate. For DTMB systems, typically f s =3.4MHz, f sym =7.56MHz, so W is equal to 3.6. Fig.3 NCO used in our Multi-GI detector. Fig.4 presents a detailed scheme of the proposed Shortened and Leakage Correlation Multi-GI (SLC-MGI) detection algorithm esponding to step to step 5, which can be explained as follows. Step (Conjugate Multiplicator). The outputs of shiftregister s 55-th unit r(n+55) and 5-th unit r(n+5) are conjugated and multiplied with the input r(n) respectively. Step 3 (Shortened Accumulator). The two multiplications of step are taken as the inputs of two shift accumulators. The shift accumulator has L registers for both PN4 and PN945. It is developed based on the relationship between R(n+) and R(n) as () and (3) shows. The shortened comes from the fact that the accumulator only utilizes a shortened copy-region with length L =65 instead of 434 for PN945, which will also be helpful to determinate the threshold value of step 5 due to the same elation length. Step 4 (Maximum Absolute). Calculate the absolute values of two elation values, R (n) and R 3 (n) ; define R( k ) and R3( k 3) as the maximum values among continuous N abselations respectively. k and k 3 denote the esponding k = arg max R n peak times, which can be expressed as ( ) and k arg max R 3 3 n n =, n [ N ], respectively. As a selected frame length, N should be larger than or equal to 475, which is the largest frame length defined in DTMB system. Step 5 (Leakage Slicer). The mode slicer works as follows: R ( k ) Th R ( k ), then the current frame head is if 3 3 PN4; else if R3( k3) Th R( k), then PN945 is determined; otherwise, PN595 is preferred, where Th is the threshold value. For the complicated elation calculation on PN595 is leaked/leaved out of the three modes of DTMB system, leakage slicer is inspired. R Fig.4 Diagram of the proposed SLC-MGI detector. R3 R3 R3( k3) R( k) R C. Complexity Analysis From Fig.3 and Fig.4, computation complexity of the proposed SLC-MGI detection algorithm can be concluded as Table I, where complex conjugate operation is reckoned as adder, and complex absolute operation is substituted by two multiplications. TABLE I COMPLEXITY OF THE PROPOSED SLC-MGI METHOD Adder Multi. Comparator register Symbol Selector Conj. Multiplicator 5 Shortened Accum. 4 33 Maximum Abs. 4 Leakage Slicer Total 8 8 6 847 From Table I, it can be seen that only a few multiplications and registers are used. In fact, the proposed SLC-MGI detector is robust to the data width of r(n). For the reason of the smallest logic utilization, only the MSB of r(n) is used for realization. Its robustness to noise and multi-path will be proved by compute simulations in Section IV.

F. Gong et al.: Multi-GI Detector with Shortened and Leakage Correlation for the Chinese DTMB System 79 As a comparison, one hand, peak detector in Mixed MMLS algorithm needs three adders and eight multiplications according to [5]; on the other hand, due to the elation length of mixed MMLS is 5, 4 complex adders are required for the local elation. Although it may be simplified by sacrificing clock resources, its complexity is much higher than our SLC-MGI detector. IV. SIMULATION RESULTS The performance of the proposed Multi-GI detection algorithm is assessed in this section. We determine the near optimum threshold by computer simulations and show its robustness to noise, multi-path, and frequency offset. A. Determination of the optimum threshold False Alarm Number of Iterations 9 8 7 6 5 4 3 TABLE II PROFILES OF BRAZIL TEST CHANNEL C/D/E Ensembles.5.5 3 3.5 4 4.5 False Alarm Number of Iterations.5.5 Attenuation (db) Delay (us).8.89 3.8.49..56.5.3.3.799..5 3.8.63.6..3 3.5 5.86.8 5.93 False Alarm Number of Iterations 4 35 3 5 5.5.5 3 3.5 4 4.5 5.5.5 3 3.5 4 4.5 (c) Fig.5 False alarm performance of the proposed SLC-MGI detector with different threshold values and GI modes configured, PN4, PN595, (c) PN945 As pointed in Section III, the leakage slicer gives a final judgment on PN4, PN595 and PN945 with the help of two elations and a pre-determined threshold Th. Th is very critical to the performance of our SLC-MGI detector. For preferable selection of Th, and three worst Brazil ensembles as Table II shows are chosen as the channel profiles used for simulation. Fig.5,, and (c) give the false alarm results of the proposed SLC-MGI detector with different threshold values range from.5 to 4.5 over /D/E channels, where 4 iterations are used for each of cases and SNR is db. It can be seen clearly that Th should be smaller than 4. for PN4 and PN945, and be larger than 3.6 when PN595 is configured. So in our algorithm, the value 4 is chosen as the near-optimum threshold. B. Performance on noise and multi-paths TABLE III PERFORMANCE COMPARASION UNDER DIFFERENT SNR OVER /BRAZILC/BRAZILD/BRAZILE CHANNELS WITH TH = 4, WHERE 4 ITERATIONS ARE PERFORMED FOR EVERY TESTING PN4 PN595 PN945 (db) // // // (db) // // // BrazilC(5dB) //53 /38/375 // BrazilC(dB) //39 /3/79 // BrazilD(5dB) /93/66 /95/4 // BrazilD(dB) /47/54 /696/34 // BrazilE(5dB) // // // BrazilE(dB) //8 // // We also present the false alarm performance of the proposed algorithm and conventional methods with typical 5dB and db over various channels. Just as Table III shows, the proposed SLC-MGI detector with Th = 4 and MSB only is robust to noise and multi-path, whereas conventional Mixed MMLS detector shows inferior performance. Note: the three values in the Table III are the false alarm numbers of SLC-MGI with MSB, Mixed MMLS and Mix MMLS with MSB, separated by /. 4 iterations are also used for each of cases. Wrong number of iterations over channel.9.8.7.6.5.4.3.. 5 5 5 3 35 4 45 5 Frequency offset (khz) PN4 PN595 PN945 Wrong number of iterations over channel 9 8 7 6 5 4 3 5 5 5 3 35 4 45 5 Frequency offset (khz) Fig.6 False alarm performance of the proposed SLC-MGI detector with various frequency offsets over channel, channel, where SNR = db, Th = 4. C. Effects of frequency offsets Fig.6 presents the proposed SLC-MGI detector s false alarm performance over frequency offset, where SNR = db, and Th = 4. 4 iterations are also used for each case. and channel are given for that they represent two extreme conditions according to our simulations. From Fig.6, it can be seen that frequency offset does not play a critical effect on estimation performance, which is a worthwhile PN4 PN595 PN945

79 characteristics over other algorithms like MMLS because MGI detector should begin to work previous of carrier recovery. Even for the worst case, a right probability better than 99% can still be assured for single trial. Further more, continuous M independent trials can be used to improve the ratio with a neglectable increment of complexity. For example, when M equals to 3, the right probability P r can be calculated as 3 Pr = Pe C3 Pe ( Pe) Even for P e =., P r can be as high as.9997, which is sufficient for DTMB receivers. For Mixed MMLS algorithm, even with good channel condition ( channel), small frequency offset (5 khz) and high SNR (db), our simulations show that large 734, 859, and 965 wrongs are found when PN4, PN595 and PN945 are deployed respectively, which are unsatisfactory in view of multi-gi detector should begin to run previous the frequency offset compensator. V. CONCLUSION Well designed MGI detection algorithms are a prerequisite for synchronization of communication systems with multiple configurable modes. In this paper, a novel SLC-MGI detection technique, which is robust to noise, carrier frequency offset, and multi-path, is proposed for DTMB systems. This technique is greatly simplified by shortening the elation length and leaking the elation calculation of PN595. According to our simulations, SLC-MGI detector with MSB only is far superior to the conventional Mixed-MMLS algorithm. IEEE Transactions on Consumer Electronics, Vol. 55, No. 4, NOVEMBER 9 REFERENCES [] Framing Structure, Channel Coding and Modulation for Digital Television Terrestrial Broadcasting System, Chinese National Standard GB 6-6, in Chinese. [] S. Tang, K. Peng, K. Gong, J. Song, C. Pan, and Z. Yang, Robust frame synchronization for Chinese DTTB system, IEEE Transactions on broadcasting, vol. 54, no., pp. 5-58, Mar. 8. [3] J. Wang, Z.-X. Yang, C.-Y Pan and L. Yang, A combined code acquisition and symbol timing recovery method for TDS-OFDM, IEEE Transactions on broadcasting, vol. 49, no. 3, pp. 34-38, Sept. 3. [4] F. Yang, J. Wang, J. Wang, J. Song, and Z. Yang, Channel estimation for the Chinese DTTB system based on a novel iterative PN sequence reconstruction, IEEE 8, pp.86-89. [5] F. Yang, K. Peng, J. Song, C. Pan and Z. Yang, Guard-interval mode detection method for Chinese DTTB system, IEEE 8, pp. 6-9. Fengkui Gong was born in Shandong Province, China, in 979. He received the M.S. degree and the Ph.D. degree respectively in 4 and 7, with the Department of Communication Engineering, Xidian University. Since 8, he has been a vice-professor of the State Key Laboratory of ISN. His research interest is communication signal processing and digital video broadcasting. Jianhua Ge was born in Jiangsu Province, China, in 96. He is now a professor and Deputy Director of State Key Laboratory of ISN at the department of communication engineering in Xidian University. He has worked at the DTV standardization as a DTV technical expert. His research interests include digital video broadcasting system, and mobile communication techniques. Yong Wang was born in Shannxi Province, China, in 976. He received the Ph.D. degree in 5. He is now a vice-professor of the State Key Laboratory of ISN, Xidian University. His research interest is digital video broadcasting system.