您的当前位置:首页基于FPGA的跳时延发射参考系统的基带数字信号处理

基于FPGA的跳时延发射参考系统的基带数字信号处理

2021-01-08 来源:小侦探旅游网
维普资讯 http://www.cqvip.com

第14卷第5期 光学精密工程 Optics and Precision Engineering Vo1.14 No。5 2006年1O月 0Ct.2006 Article ID 1004—924X(2006)05—0876-07 基于FPGA的跳时延发射 参考系统的基带数字信号处理 王易因 ,Rene van Leuken ,Alle—Jan van der Veen ,曾晓洋 ,章倩苓 (1.复旦大学专用集成电路与系统国家重点实验室,上海200433; 2.荷兰代尔夫特理工大学,电子工程系,荷兰) 摘要:超宽带(UWB)通信通过传输能量极低的极短脉冲和占用超大带宽来完成短距离的高速通信。由于其极高的数据 传输率,抗多径衰落,低成本和低功耗等诸多优点,成为短距离无线通信系统的研究热点。在有诸多优点的同时,它的 实现又面临着很多挑战,比如系统采样率极高,对同步精度要求极高,信道估计极为复杂。而跳时延发射参考(Delay Hopped Transmitted Reference--DHTR)系统是超宽带系统的一个低复杂度的实际可行的方案:通过在模拟域做相关操 作,大大降低了系统的采样率和数字信号处理的数据量;把传输的参考脉冲作为信号脉冲的污损模板,绕过了复杂的信 道估计问题。本文的工作是AIRLINKl4 项目演示任务的一部分。文中说明了DHTR UWB通信系统的工作原理;分析 给出系统的框架结构;通过对文献[3]给出的接收信号数学模型进行简化,分析得到了同步解调的方法;并针对DHTR 系统符号同步方法,提出了两种电路实现的结构一串行同步和并行同步结构;然后用FPGA实现了DHTR系统发射端 和接收端的基带数字信号处理部分;最后搭建了一个完整的验证平台,对系统进行了功能验证。此系统的电路综合结果 表明系统实现了低复杂度、低成本,证明了它在超宽带通信中的可行性和实用性。 关键词:高速通信;跳时延发射参考;超宽带;同步电路 文献标识码:A 中图分类号:TN925 DSP implementation for delay hopped transmitted reference system based on FPGA WANG Yi~yin ,Rene van Leuken ,Alle—Jan van der Veen ,ZENG Xiao—yang ,ZHANG Qian-ling 1.State Key Laboratory of ASIC and System,Fudan University,Shanghai 200433 China; 2.Department of Electrical Engineering,Delft University of Technology,Holland) Abstract:Ultra Wideband(UWB)system iS an attractive high speed wireless communication scheme for short range corn— munication.It transmits extremely short pulses,which occupy ultra wide band.It gets more and more attention,because of its high data capability,fading resistance,low cost and low power,etcWhile having SO many advantages,it also con— .fronts many challenges,such as very high system sampling rate,rigor request for synchronization and formidable channel estimation.Delay-Hopped Transmitted Reference(DHTR)system is a low complexity and practical scheme for UWB communication.It drags down the sampling rate of the system by dealing with correlation in the analog domain and by— passes the formidable channel estimation by using the reference pulse aS a dirty template for the signal pulse.The work shown in this paper is part of the demonstration of AIRLINKⅢproject.The paper explains the working principle of the Received date:2006-03-16;Revised date:2006-07-27. Foundation item:Supported in part by the Dutch Min.Econ.Affairs/Min.Education Freeband Impulse AIRLNK Project. 维普资讯 http://www.cqvip.com

No.5 WANG Yi-yin,et al:DSP implementation for delay hopped transmitted reference system…… 877 DHTR system and proposes the system architecture.It shows the synchronization and detection method by analyzing and simplifying the signal model proposed in reference[3].Two circuit architectures for synchronization are proPosed:the se— rial solution and the parallel solution.Then the digital signal processing parts of the DHTR system are implemented on a Field Programmable Gate Array(FPGA).A complete verification platform is constructed to verify the function of the sys— tem,The 1ow complexity and 1ow cost implementation of the whole system proves its possibility and practicability in UWB communications. Key w0rds:hi speed wireless communication;Delay Hopped Transmitted Reference(DHTR);Ultra Wideband(UWB); synchronous circuit 1 Introduction For overlay on existing frequency alloca— tions,along with promises of high data rates, low cost and low complexity,the ultra—wideband f UWB)is an attractive technology for wireless communication.A practical UWB communica— tion scheme is given by the Delay—Hopped Transmitted—Reference communication system (DHTR system)proposed by Hoctor and Tom— linsonE 一 .It is based on the transmission of Dairs of pulses whose correlation carries the in— formation:this is unchanged after convolution by the propagation channel since both pulses ex— Derience the same distortion. For synchroniza— tion and detection,the individual channel coeffi— cients do not have to be estimated,which makes this a much more attractive scheme than some of the proposed rake transceivers. 0ur aim in this paper is to consider a practi— ca1 implementation of the digital parts of the DHTR system on an FPGA prototype board・ We propose two hardware architectures for de— tection and synchronization:a serial and a paral— lel architectures.We will show the synthesis re— suIts for the serial architecture:this gives an in— dication on how fast the transceiver can be and .how many resources it employs. This research is part of the AIRI INK pro— ect[4]where other work packages consider the antenna design,analog electronics,and commu— nication/networking layers.In the design,the relative1v low clock speed of an FPGA is offset by its high degree of parallelism and I/0 capabil— ities,so that nonetheless an acceptable data rate can be achieved. The paper is organized as follows.In sec— tion II,we introduce the structure of the DHTR svstem,its working principles,and explain how to mode1 a simplified version of the system.In Section III,algorithms for synchronization and demodulation are presented.We propose two ar— chitectures to implement the algorithms.Section IV shows the synthesis results for the serial ar— chitecture.Finally,conclusions are drawn in Section V. 2 DHTR system The delay—hopped transmitted—reference communication system transmits pulses in pairs (as a doublet).The first pulse is used as a refer— ence and the second is used to carry the informa— tion.The pulses are separated by a short time interva1,which is known by both the transmitter and the receiver in advance.This separation changes from doublet to doublet according to a user—spec:ific“delay code”.The analog part of the receiver correlates the received signal with several time shifts using a bank of delay lines, integrates the results,and subsequently samples the outputs for digital synchronization and de— modulation.Thus,the analog parts of the sys— tem do not contain any time—dependent or para— metric parts,which is important since they run at maximum speed. The transmitted messages are r epresented by symbols.A symbol is composed of several doublets,where,similar to CDMA systems, 维普资讯 http://www.cqvip.com

878 Optics and Precision Engineering Vo1.14 each doublet represents a chip.Fig.1 shows an example of the signal pattern to transmit one symbo1.In the design of our system,a symbol s consists of 8 chips cEk],k一1,…,8,where each chip is a frame with a duration of 2O ns.In the frame,two narrow pulses g( )form a doublet d ( ).The first pulse is a fixed referenee pulse。 and the second one has a polarity which depends on the symbol 5 and is modulated by cEk].The doublet can be formulated as ( )一g( )+cEk]*s g( 一D[是]),k一1,…,8. (1) Where,s is the symbol value and can be——1 or 1;cEk]is the chip value,which can also be一1 or 1,and constitutes a user—specific code.D[是] is the user—specific delay time sequence,with values that are a multiple of 0.5 ns.In our sys— ,tem,we use 5 possible values,D[是]∈{0.5 ns, …,2.5 ns}. One symbol Fig.1 Structure of one transmitted symbol A block diagram of the transceiver system is shown in Fig.2.The upper part of Fig.2 in— dicates the transmitter and the lower part depicts the receiver.The pulses transmitted by the an— tenna in the transmitter go through the wireless channel and are received by the antenna in the receiver.The dashed line divides the digital part from the analog part of the transmitter and the receiver.All the digital parts are implemented on an FPGA. Twelve output pins of the FPGA are used to represent the two pulses in a frame and their displacements.In particular,binary pins Po, …,P5 represent the presence of a positive pulse at displacements of 0,…,5 times 0.5 ns。and similarly N0,…,N5 represent the presence of negative pulses.If there is a signal to transmit, P[O P【1 P[2 P[3 P[4 P[5 《 N0 l广_] M1 N2’ N3 N4 N5 Analog part Fig.2 Schematic block diagram of DHTR system then either P0 or No will be logically‘1’to re— present the reference pulse,and only one of P1, …,P5,Nl,…,N5 will be‘1’to represent the signal pulse and its time offset.The time shifts are implemented by analog delay lines(D1,D2, …,D5 in the figure;in practice,we use a slightly different scheme with only a single tapped delay line).The delayed signals are add— ed together and sent into the pulse generators (one for a positive pulse and a separate one for a negative pulse),which generate narrow pulses in sequence.The analog pulses are added to— gether,sent into the amplifier and transmitted by the antenna.The pulses are transmitted and then received by the antenna in the receiver.The received signals go through a bank of delay lines (the same delay periods as used in the transmit— ter),are correlated,integrated over a period of 20 ns,and sampled by an A/D converter.This procedure gives a strong positive(or negative) response in those delay branches that match the delay of the transmitted doublet,and approxi— mately a zero response for other(non—matching) delays.The digital samples are sent to the digit— al part of the receiver implemented on an FPGA 维普资讯 http://www.cqvip.com

No.5 WANG Yi-yin,et al:DSP implementation for delay hopped transmitted reference system…“ 879 for synchronization and demodulation. To be able to test the digital part of the re— ceiver system,we have also implemented a sim— pie channel emulator,running on an FPGA,as shown in Fig.3.It has inputs P0,…,P5,No, …,N5 as generated by the digital part of the transmitter, and generates the corresponding outputs X1,…,X5 suitable for the digital part of the receiver.The implemented emulator uses an ideal response: X Ek]一A *cEk]*s A ≥O; k一1,…,8;DEk]∈{0.5 ns,…,2.5 ns}; —DEk]/O.5; ∈{1,…,5}. (2) where A is a gain parameter which depends on the transmitted delay index i and the received delay index J,and is related to a channel correla— tion coefficient.Ideally,if i—J,A —A,else A 一0[3]. |D l P[I l 朋 P 2 P[2 |D 3 P[3 船 P 4 H4 p[5 P[5 symbol No NO Ml Ml N2 M2 324 N3 M3 N4 M4 筋 325 M5 M5 FPGA FPGA FPGA Transimitter C:hannel emulator Receiver Fig.3 DHTR system emulator 3 Receiver synchr0nizati0n and de— modulation 3.1 Detection and synchr0nizati0n At the receiver,the user code and delay time sequence are known information and they are repeatedly used for every symbo1.The first thing to do at the receiver is to synchronize,i. e.,to detect whether we have received a valid user signal and to find the position of the first chip of a symbol,taking into account an un— known integer delay.The detection is done by matching the desired user code(chip code and delay code)with a single symbo1.Every symbol is composed of 8 chips,and since each chip cor— responds to a sample,there are 8 possible posi— tions for the first chip.We have to check them all to find the best match.which corresponds to a maximum correlation with the code.If a mes— sage was transmitted,then for each sample,one of X】,…,X5 will have a value of+A or—A, namely the output corresponding to the trans— mitted delay between the pulses of that frame.If we are synchronized。then for the志一th chip,we choose X,[是]according to the user specified de— lay time sequence,J=D[k]/O.5,multiply with the corresponding chip value cEk],and sum the results over 8 chips:this gives a matched output of r一8 As.From r,we can estimate A as 1/8  Jr 1.If we are not synchronized,the samples will not add coherently,and r and the estimated A will be a small number.Thus,at the proper synchronization position,we will get the maxi— ilium estimated A.To synchronize,we check all 8 possible offsets,get 8 estimated values for A and find the maximum one--max—A,and the corresponding position--max——index. To test whether there was a signal at all, we need to compare the maximum estimated A to a threshold value.This threshold value can be determined by analyzing the variance of the value that we will obtain in the case of noise—only.Af— ter choosing a desired false alarm level,it can be determined with the help of statistical signal processing theory.This would require knowl— edge of the noise power.In the absence of this information,we use the average value of the es— timated A over all possible positions as the threshold.If the maximum estimated A is n times larger than the average,we decide that we have received a desired user signal,otherwise it is j ust noise.Again,the correct a should be de— termined using statistical signal processing theo— rv.In the text below,we use d一3. After svnchronization,we can demodulate 维普资讯 http://www.cqvip.com

880 Optics and Precision Engineering Vo1.14 the symbo1.In fact,the algorithm for demodu— lation is similar to the algorithm for estimating A,since it requires the correlation sum r.At the correct position,the receiver starts to demodu— late.It chooses X,[k]according to the user specified delay time sequence,multiplies it with the user code and accumulates the products.Af— ter accumulation for 8 chips,the sign of the sum is the demodulated symbo1.The demodulation is described by the following pseudo code: r一0: for(忌一0;忌<8;k++) J—DEk]/O.5; r—r+X,Ek]*cEk]; symbol—sign(r); 3.2 Serial synchrOnizatiOn architecture Following the algorithm described above, we propose two FPGA implementations struc— tures for svnchr0nizati0n.The first one is a seri— al architecture.The 8 possible offset positions are checked in a serial way.For each offset,the correlation calculations are the same.Since they do not happen concurrently,the corresponding resources can be reused which saves a lot of area.The operations are assigned to every cycle according to the algorithm.Fig.4 shows the ar— rangement for all the cycles.In the first cycle, the first sample read in is used as the starting point.The receiver reads in a new sample every clock tick of 2O ns.As shown in Fig.4,the numbers in the first row of every group is the cycle range.The numbers in italic represent the sequence of each batch of 8 samples.The num— bers in bold represent the sequence of 8 chips we use to compute r and estimate A.Each batch of 8 chips starts at a different offset position,by skipping one sample in between.The numbers in the right column show the corresponding start position of the batch in black referred to the batch in italic.For example,in the row of the second batch of 8 samples in italic,the bold 0 corresponds to the italic 1.At this time we check position 1 as the start point.The stars* are the skipped samples in order to have some offset to check different positions. At the clock tick corresponding to a skipped sample。we have time to(i)calculate the abso— lute value of the estimated A,(ii)accumulate the value for the computation of the mean,and (iii)compare the absolute value with max—A, which is the record of the maximum A seen so far.If the new A is larger than this maximum, we update max_n and max_index.The compari— son and the accumulation can be done concur— rently because they use different function re— sources without any data dependency.For each batch of 8 samples,an accumulation is done to get the estimated A and every cycle there is an addition operation.To balance the operation of every cycle,we can schedule the calculation of 3 *avg—A,which is used as a detection thresh— old,at cycle 73 as shown in Fig.4. 1—8(cycles) Possible posmon 0 1 2 j 4 5 6 7 0 1 2 3 4 S 6 7 0 0 1 2 j 4 5 6 7 l 十0 1 2 3 4 5 6 Updatethemax A \===二二二/and一 - Accumulate !: absolute value es石mate 57.64 ・ 0 1 2 j 4 5 6 7 7 2 3 4 5 6 7十0 Get estimatedA —————— 一 ● 65.72、 . .——————. . 0 1 2 j 4 5 6 7 l 2 3 4 5 6 7十 —————— ————__/ Get avg3A 73.80 nad 0 J 2 j 4 5 6 7 L0lnpare 十 Imax A1wihtIavg3A Fig.4 Serial synchronization:arrangement of cycles At cycle 7 3,we decide whether the signal was detected.If not,we continue another scanning round,otherwise we skip a specific number of samples to synchronize to the computed offset position(max_Index)and start to demodulate. For example,if the proper position is 7,then we skip 7 samples and start demodulation at cycle 8O.In the most ideal case,74 cycles are needed for svnchr0nizati0n. A disadvantage of this scheme is that syn— 维普资讯 http://www.cqvip.com

No・5 WANG Yi-yin,et al:DSP implementation for delay hopped transmitted reference system…… 881 chronization requires 9 symbol periods,even if effectively only a single symbol period is used to detect the signal:this is not efficient from a sig— nal processing point of view. 3.3 Parallel synchr0nizati0n architecture Since FPGAs have abundant resources。an alternative architecture is a parallel one.We can check all 8 possible offset positions concurrently and thus reuse the same samples for different positions.Fig.5 shows the function blocks for this kind of receiver.For each possible offset position i,there is a function block P,( 一0,…, 7)that operates on the samples starting with a corresponding offset,as triggered by a time— shifted‘start’signa1.Every P block calculates the average estimated A(signal A in the fig— ure),demodulates the sample(signal S ),and updates the average value of the estimated A ev— ery 160 ns(8 chips or 1 symbol period)at con— secutive moments.These estimated A’s are compared to a threshold value to detect whether there was a signal and are also used to find the correct offset position.A selector function block selects the corresponding signa1.This is the idea to do the synchronization using a parallel archi— tecture.The serial method uses fewer resources. but takes a longer time to synchronize and is not efficient from a signal—processing viewpoint.The parallel method uses more resources,but is fas— ter in synchronization:in fact it wil1 detect the beginning of a packet as soon as it arrives. 4 Resuhs In order to test the functionality of the DHTR system.we encapsulate it into a VHDL soft—core of an Atmel AVR micro—controller en~ vironment(shown in Fig.6).The whole envi~ ronment is implemented on a Xilinx Spartan3 (xc3s1500fg676—4)FPGA.A C-program。run~ ning on the AVR,enables communication be~ tween a user and the DHTR system.The user inputs a message using the keyboard.The mes~ Fig.5 Receiver using parallel synchr0nization archi tecture sage is transmitted to the AVR,translated into symbols by C program and sent to the transmit— ter hardware.The demodulated symbols from the receiver are collected by the AVR.They are translated into a received message and shown on the screen. Fig.6 Architecture of testing platform The transmitter,the channel emulator and the receiver with the serial svnchronization ar— chitecture are implemented on the FPGA as a Wishbone client.The total number of lines of the VHDL description for this DHTR system is less than 1 5OO.The synthesis results of the sys— tern with the serial svnchronization architecture are shown in Tab.1。As seen in the table,the system with the serial synchronization employs only a few resources,as indicated by the number 维普资讯 http://www.cqvip.com

882 Optics and Precision Engineering Vo1.14 of equivalent logic gates.As mentioned above,a disadvantage of this scheme is its slow acquisi— 5 Conclusions This paper presents the design of the digit— al part of a DHTR UWB communication system. tion time:it will take at least(c+1)*c+1 (where c is the number of chips per symbo1) times the cycle time to complete one scanning It is a mixed signal system.The transmitter modulates the symbols according to a user speci— led code:a polarifty and a delay time sequence. round.As c increases,this time increases qua— dratically.This is not desirable from a signal— processing viewpoint.From these numbers in Tab.1,we can extrapolate thatthe number of .The major challenge in this design is to obtain svnchronization at the receiver.Two ways are proposed to do synchronization:the serial meth— od and the parallel method.The serial method uses fewer resources,because it can share the resources used in the parallel receiver will be a— bout f*233 gates.The time spent in one syn— chronization cycle is c+(c一1).When c increa— ses。the area and the acquisition time both in- crease linearly.In most cases,this is a better solution.According to the specific application, calculation resources when checking all possible positions in serial,but it takes a relatively long time to synchronize.The parallel method uses mote resources.because it checks all possible offset positions concurrently,but it is therefore we can combine these two architectures to make a tradeoff between the acquisition speed and the chip area. able to achieve synchronization within a single Tab.1 Synthesis results of DHTR system with serial synchronization architecture symbol period.The design is targeted towards implementation on an FPGA development boards with a Xilinx Spartan3 device.We build a com— plete test environment to verify our design.The test results proves the functionality of the design is correct. RefeFences: NSON H.Delay-hopped transmitted—reference RF communication[C]. IEEE Conference on [1] HOCTOR R。TOMLIUltra Wideband Systems and Technologies,2002:265—269. NGER A,WEI 1 ES K.Delay hopped transmitted reference experimental results[C].IEEE E2] STRALEN N,DENTICon ference on UItra Wideband Systems and Technologies,2002:93-98. gnal processing model for a transmit—reference UWB wireless communi— [3] TRINDADE A,DANG Q H,VEEN A J.Sication systemiC].IEEE Con rence on Ultra Wideband Systems and Technologies,2003:270—274. [4] http://www.airlink.tudelft.nl[Z]. Brief professional biography of the author:WANG Yi—yin received her B.S.degree in Electrical Engineering from Fudan University in 2002,M.Sc.degree in Microelectronics from Delft University of Technology in 2005 and M.S.degree in Microelectronics from Fudan Universi— tv in 2006.Now she is a research assistant in Fudan University.Her current research interests include wireless communication system,especially UWB sys— terns。high level synthesis and VLSI circuit design.E—mail:yiyinwangfd@hot— maj1.com 

因篇幅问题不能全部显示,请点此查看更多更全内容