汽車防碰撞預警執(zhí)行系統(tǒng)的設計,汽車,碰撞,預警,執(zhí)行,系統(tǒng),設計 Pulsed laser ranging techniques based on digital signal processingmethods for automobile anti-collision applicationZhihui SUN * , Jiahao DENGSchool of Aerospace Science and Engineering, Beijing Institute ofTechnology,No.5 Zhongguancun South Street, Beijing, China, 100081ABSTRACTA 1.55 m digital laser radar system is designed and implemented for automobileanti-collision application. In order to reduce the influence of foggy, rainy and snowyweather on laser detection, digital signal processing methods are adopted.Multi-pulse coherent average algorithm is used to improve the signal-to-noise ratioof echo by N times. The correlation detection algorithm is adopted to estimate thetime-of-flight. Multi-time delayed correlating method is used to improve thetime-of-flight estimation resolution. Experimental results indicate that the digitalsignal processing methods in this paper can reduce the influence of bad weatherconditions, and obtain high range accuracy.Keywords: automobile anti-collision, distance measurement, laser ranging, laserradar, time-of-flight estimation,digital signal processing, weak signal detection,correlation detection1. INTRODUCTIONTraffic accidents take place frequently with the increase of automobile number andspeed. Driving safely draws more and more attention and the research onautomobile anti-collision system becomes hot. Automobile anti-collision systemsadopt mm-wave radar or laser radar to detect proceeding vehicles, obstacles,pedestrians and measure the distance of objects. When the distance is less than thesafety distance, the systems alert the driver or brake automatically. Therefore, drivesafety in poor weather condition such as rainy, snowy, foggy is enhanced and trafficaccidents can be avoided. Compared with mm-wave radar, the main advantages oflaser radar are more mature, reliable and cheaper. However, the disadvantages oflaser radar are its inability to penetrate rain and fog, and also there is the questionof eye safety regarding using high power pulse lasers, so laser radar doesntencounter the favorable opinion of car makers for a long time. Recently, along withthe progress in manufacture technology of laser diode, photodetector, and alsosignal processing techniques, the weakness of automobile laser radar is overcome.Some companies such as IBEO, Omron automotive electronics, Daihatsu havedeveloped advanced automobile laser radars and their performances are as good asmm-wave radar1-3In this paper, a 1.55 m digital laser radar system is designed and implemented forautomobile anti-collision application. By comparing 1.55 m laser with 0.9 m laser,the 1.55 m laser is safe to eyes, and its capabilities in detection and penetrating fogare stronger. The system adopts high speed analog to digital converter (ADC) tosample the pulse echo signal, and then digital signal processing methods are usedfor signal preprocessing and time-of-flight estimation. Experimental results showthat the detection capability of weak echo signal is enhanced, and the rangingaccuracy of the system is improved. Thus, the performance of digital laser radarsystem in poor weather conditions is improved.2. OVERVIEW OF AUTOMOBILE LASER RADAR SYSTEM2.1 Working principleThe working principle of automobile laser radar system is illustrated in Fig 1. Whenthe range to a preceding vehicle R is less than S1(alarming range), the system alertsthe driver to slow down; when the range R is less than S2(braking range), the systemalerts the driver to brake or brake automatically.Fig 1. Working principle of automobile laser radar systemThe range to a preceding vehicle R is calculated on the basis of elapsed timebetween transmission of a laser pulse andreception of reflected light4. Knowing the laser pulse travels at the velocity of thelight c (m/s) and measuring the time-of-flight of the laser pulse t, the range R (m) isgiven byR=ct/2(1)The laser radar range equation is the foundation for designing the system andevaluating the performance of the system. Given that the target is larger than thebeam and has a Lambertian reflectance distribution, the equation is5Pr=?(2)WherePr= power received in wattsPt= power transmitted in wattst= transmitting optics efficiencyr= receiving optics efficiency= reflectanceD = entrance pupil diameter in metersa= atmospheric transmission factor (one way) = exp(-R),where = atmospheric attenuation coefficient in km-1R = range in metersAs can be seen from Eq. (2), several factors influence the performance of laser radarsystem. Generally, when do research on the performance; the laser radar rangeequation is often expressed in the form of signal to noise ratio (power):SNR =(?t?)2=(?t?)2(?th?t?)2(3)where NEP is noise equivalent power in watts, and it is interpreted as the standarddeviation for the Gaussian distribution of the additive noise. NEP is combinedtogether by the noise in detector and preamplifierNEP= ?t? ?t?h?h?(4)WhereNEPdetector= noise equivalent power of detectorNEPpreamplifier= noise equivalent power of preamplifierEq. (3) relates the signal to noise ratio (SNR) to range for given hardware parameters,weather conditions, and target characteristics. SNR is an important parameter inevaluating the performance of the laser radar system.Probability of detection is another important parameter, based on pulse detection inwhite noise using a matched filter, the probability of detection isPd=?+?erf(? ?)(5)WherePd= probability of detectionerf (x) = unilateral error functionTNR = threshold-to-noise ratio, is expressed as follows:TNR= ? ?t?t(6)Where = pulse width, FAR = average false alarm rate = Pfa. PRF,and Pfa= single pulse false alarm rate, PRF = laser pulse repetition frequency2.2 Comparison between 0.9 m and 1.55 m laser radarsGenerally, 0.9 m laser radars exhibit good performance only with cooperativeobstacles in good visibility conditions; in order to detect non-cooperative obstaclesand improve performances in poor weather conditions, devices outside Class I arerequired6, and the problem of eye safety is caused. Compared with 0.9 m laser,1.55 m laser is safe to eye, and the capability of penetrating fog is stronger;consequently the allowed transmitted laser pulse energy is increased, and thedetection capability in low visibility is improved.As can be seen from Eq. (2), the atmospheric effects limit the performance of thelaser radar system. Two-way atmospheric transmission factor TaisTa=?=exp(-2R)(7)where atmospheric attenuation coefficient is7=?th?(?thh)-q(8)where = laser wavelength in mRv= visibility distance in kmq = the size distribution of the scattering particles=1.6for high visibility (R v 50 km)=1.3for average visibility (6 km R v 50 km)=0.585V1/3for low visibility (R v 6 km)Fig 2 shows a plot of two-way atmospheric transmission factor versus range for 0.9and 1.55 m lasers, as the visibility distance Rvis 0.5 km, which stands for moderatefoggy weather condition. As can be seen from Fig 2, 1.55 m laser is stronger in thecapability of penetrating fog than 0.9 m laser.Comparison between 0.9 and 1.55 m laser radars in detection capabilities is made,to detect a low reflecting pedestrian with a high probability (Pd=0.999, Pfa=10-13)in moderate foggy weather condition (visibility distance Rv=0.5 km). As laser pulsewidth =100 ns, laser pulse repetition frequency PRF =10 KHz, from Eq. (5) and Eq.(6), the required SNR is calculated, SNR 12 dB. As is shown in Fig 3, the horizontalline is the required SNR to achieve the high probability.Si-APD and InGaAs-APD are used as detectors respectively by 0.9 and 1.55 m laserradars. The noise equivalent power of APD (NEPdetector) is8NEPdetector=NEPHZ?(9)where NEPHzis noise equivalent power of APD in watts perHz , and B is noisebandwidth in hertzThe noise equivalent power of preamplifier (NEPpreampl) is9NEPpreampl=?t(10)wherek = Boltzmanns constantT= the temperature in degrees KelvinN = noise factor of the preamplifier Res= the responsivity of APDRL= the load resistor =1/2BC, and B = noise bandwidth in hertz,C= capacitance of the APDFrom Eq. (3) (4) (8) (9) (10), the SNR of 0.9 and 1.55 m laser radar system can beobtained with following parameters:Pt= 50 W; t= r= 0.6; = 0.15(0.9 m) or 0.25 (1.55 m) based on pedestrian astarget; D =0.04 m; R v =0.5 km;k =1.3810 -23 J / K ; T=295 K (22C); N=2; C= 1 pF; B=35 MHz;NEPHz=10 -14 W/Hz ( Si-APD) or 0.1510 -15 W/Hz (InGaAs-APD)Res=9.4 A/W (Si-APD) or 9 A/W (InGaAs-APD);Fig 3 shows a graph of SNR versus range for 0.9 and 1.55 m laser radar systems asthe visibility distance R v =0.5 km. With the probability (Pd=0.999, Pfa=10-13),required SNR 12 dB; as can be seen from the graph, the maximum detectionrange of 0.9 m laser radar system is about 150 m, while 200 m for 1.55 m laserradar system. The theoretical comparison results show that the detection capabilityof 1.55 m laser radar is stronger in poor weather conditions. So, development of1.55m laser radar can improve performance in low visibility such as fog conditions,and if advanced signal processing methods are adopted, the performance can beimproved further more.2.3 1.55 m digital laser radar system constructionAs shown in Fig 4, 1.55 m digital laser radar system consists of following threecomponents: transmitter that drives pulse reference signal and emits pulsed laserlight; receiver that condenses the reflected light, undergoes photoelectricconversation and weak pulse signal amplification; signal processing system thatsamples the pulse echo signal by high speed ADC and undergoes signalpreprocessing and time-of-flight estimation by digital signal processing methods.The signal processing system is based on field-programmable gate arrays (FPGA) anddigital signal processor (DSP). FPGA is used to complete time sequence controlfunctions such as laser pulse reference signal generation, high-speed ADC sampling,data buffering and interrupt signal generation. DSP is used to implement signalpreprocessing and time-of-flight estimation algorithm.Working principles of the 1.55 m digital laser radar system are as follows: DSPstarts FPGA to generate a pulse reference signal with 100 ns pulse width and 10 KHzrepetition frequency, and laser driving circuit amplifies the pulse reference signal tocontrol the diode laser to emit pulsed laser light, then transmitting optics shape thelaser light into narrow beam and transmit forward. Receiving optics condense thelight reflected back from the reflecting object, and the photodetector converts it toan electrical current pulse signal , then transimpedance amplifier converts the weakcurrent pulse signal to a voltage pulse signal , the variable gain amplifier furtheramplifies the voltage pulse signal suitable for the input voltage range of ADC. At thesame time of generating the laser pulse reference signal, the ADC samples the pulseecho signal at 200 MHz equivalent frequency under the control of FPGA, and FPGAstores the data in its inner random access memory (RAM), when required data havebeen sampled, FPGA interrupts DSP; DSP responses the interruption and reads thedata in, and implements multi-pulse coherent average algorithm to increase the SNRof pulse echo signal, then adopts correlation detection method to estimatetime-of-flight, and further improves the resolution of time-of-flight estimation bymulti-time delayed correlating method.3. PULSE ECHO SIGNAL SAMPLINGDual-channel ADC with 10-bit resolution and 100 MHz sampling frequency isadopted. The pulse echo signal is sampled alternately by channel A and B under thecontrol of reversed clocks, and the equivalent sampling frequency is doubled to200 MHz. The working principle is shown in Fig 5.Alternate sampling is strict with the time sequence of clocks, and the reversed clocksignal of alternate sampling is generated by FPGA. Then the data sampled are storedin inner RAM of FPGA. Principle of sampling control and data buffering by FPGA isshown in Fig 6, and the process is implemented by Verilog-HDL languageprogramming. As shown in Fig 6, input clock frequency of FPGA is 50 MHz; two 100MHz reversed clocks are generated by on-chip phased-lock loop (PLL). The clocksand data of dual-channel ADC are connected to ADCLOCKA, ADCLOCKB, DBA, andDBB respectively. The data sampled are stored in two RAMs of FPGA. Becausedual-channel ADC operates alternatively, the sampling data should be recombinedby the Bus Controller. When the bus address is even, the data of channel A areoutput; when the bus address is odd, the data of channel B are output.4. MULTI-PULSE COHERENT AVERAGEAs can be seen from Fig 2 and Fig 3, despite performance of 1.55 m laser radar infoggy weather is better than that of 0.9 m laser radar, as the range increase,atmosphere attenuation becomes severe, pulse echo signal is weak and sometimeseven submerged in noise. In order to increase the SNR of pulse echo signal,multi-pulse coherent average algorithm is adopted. Basic principle of multi-pulsecoherent average algorithm is: multiple pulse echoes are sampled by high speedADC, and then the sampling values are accumulated corresponding to their relativepositions.The pulse echo signal can be expressed as follows:X(t)=As(t)+w(t)(11)wheres (t) = normalized pulse signalA= the amplitude of pulse echo signalw (t) = zero mean Gaussian white noise and its root mean square value is N pulse echoes are sampled ,if there are M sampling points in each echo andsampling interval is t, then the value of sampling point j (j=0,1,M-1) in pulse echoi (i =0,1,N-1) isx(? t?)=As(? t?)+w(?+jt)(12)Where tiis sampling start time of pulse echo i, and the sampling start time ofdifferent pulse echoes is required as the same, Eq. (12) can be abbreviated as:xij=Asij+wij(13)M sampling value of a pulse echo is stored and summed respectively with Msampling value of last pulse echo, when N pulse echoes have been sampled andaccumulated, the coherent average value of point j is?t?t?t?(14)Eq. (14) can be further arranged as?t?t? ttt?htInput SNR ?powert is define asSNRit?6tThen output SNR ?powert isSNR?N SNRi?7tAs can be seen from Eqt ?7t, the SNR of pulse echo signal is improved by N timeafter processed by multi-pulse coherent average algorithmt Experimental resultsof multi-pulse coherent average algorithm are shown in Fig 7; ?at is a plot ofpulse echo signal sampled by high speed ADC, and the noise is high; ?bt, ?ct and?dt are the processing results when coherent average times N?, h?, ?t Ascan be seen from Fig 7, the SNR of pulse echo signal is improved gradually as thecoherent average times increaset However, the processing time becomes longeras the coherent times increaset So the choose of coherent average times is thetradeoff between SNR improvement and processing timetht ht TIME-OF-FLIGHTTIME-OF-FLIGHT ESTIMATIONESTIMATIONht?ht? CorrelationCorrelation detectiondetectionAs can be seen from Eqt ?t, Target range is calculated on the basis oftime-of-flight estimationt And the ranging accuracy mainly depends on thetime-of-flight estimation accuracyt There are several time-of-flight estimationmethods: leading edge detection, zero-crossing detection, peak detection,constant-fraction detection and correlation detection ?-? t Compared withother detection methods, correlation detection is the best in performance andless affected by the noise? t In this paper, correlation detection method isused to estimate time-of-flighttThe mathematical model of pulse echo signal can be expressed asX?ntAs?n -?t?w?nt? ? ? ? ? ?8twhere n?is the sampling point at time delay, M is the sampling length, and otherparameters are the same as Eqt ?ttThe correlation detection method uses the position of peak of cross-correlationfunction r?kt as the estimated? ? ? argmaxr?ktr?kt?t?tt? t?htThe range is calculated byR? ? ?twhere ?t is time-of-flight estimated, ? is the sampling intervaltAs can be seen from Eqt ?t, the ranging resolution by correlation detectionmethod depends on the sampling frequency of ADC, and high samplingfrequency provides better range resolutiont In this paper, ? MHz ADC is used,the time-of-flight estimation resolution is h ns, and the ranging resolution is ?t7hmt In order to improve ranging resolution further, adopting ADC with highersampling frequency is a choice, but the problem is that the circuit becomescomplicated and hard to implemented, also the cost is hight So this paperpresents multi-time delayed correlating method to improve ranging resolutionwithout using higher sampling frequency ADCtht?ht? Multi-timeMulti-time delayeddelayed correlatingcorrelating methodmethodTime-of-flight estimation by correlation detection method is to calculate theposition of cross-correlation peakt When time-of-flight is integer times ofsampling interval ?, the cross-correlation peak can be sampled; otherwise, theneighbor of the peak is sampled, and time-of-flight estimation error is producedtIf the cross-correlation peak delays a certain time so that the peak can besampled, then the certain time delayed can modify the time-of-flight estimationerror caused by samplingt The delayed cross-correlation function isr?k?dt?t ? t? ? ?t?t ? t ? ? ?twhere d is a certain time delayed, and it is less than the sampling intervaltAs can be seen from Eqt ?t, to calculate delayed cross-correlation function, justneed to delay the reference signal for a certain time d, and then calculatecross-correlationtIf let sdnsn-dand rd?ktr?k?dt,then Eqt?tcan be expressed as follows:rd?kt?t ? t? ?twhen d?,?,?, ,?,the delayed reference signal ist? ,t? t? , ,t?,as shown in Fig 8tThe delayed cross-correlation functions are? ?t?t? ? ?t ? t? ? ?t ? t? ?tIf the maximum of delayed cross-correlation functions is ?t? i ? ? ?t,then the time-of-flight estimated t ist? ?a?tAs can be seen from Eqt ?t, the time-of-flight estimation resolution isimproved by K times on the basis of correlation detection methodt In this paper,five delayed reference signals are used, and the time-of-flight estimationresolution is improved by h timest When sampling frequency of ADC is ? MHz,the sampling interval ?h ns, then time-of-flight estimation resolution improvedby multi-time delayed correlating method is ? ns, and the ranging resolution isimproved to ?t?h m correspondinglyt6t 6t CURRENTCURRENT EXPERIMENTEXPERIMENT RESULTSRESULTSCurrently, indoor short range measurement experiments have been done toverify the multi-pulse coherent average algorithm and the time-of-flightestimation algorithm, and the target is white wallt As shown in Fig h, the resultsare displayed in Texas Instruments DSP integrated development environmentCCSt In each graph, the upper is the laser pulse echo signal sampled by ADC; themiddle is the result of multi-pulse coherent average algorithm; the lower are theresults of time-of-flight estimation and range calculationt In Fig h, from the left tothe right, the true range is ?th m, ?t? m, ?th m and ?t? m respectively; and thecalculated range is ?t?h m, ?t?h m, ?t7hm and ?th m respectively; the maximumrange error is ?t?h mt Also the algorithms adopted decrease the effect of pulseecho amplitude and shape variations on ranging accuracyt7t 7t CONLUSIONCONLUSIONIn this paper, a ?thh m digital laser radar system is designed and implemented,and digital signal processing methods are used in pulsed laser rangingtechniques to enhance the detection capability of weak signal and improveranging accuracyt The basic working principles of automobile laser radar aredescribed, comparison results of ?th and ?thh m laser radars show that thelatter has better performance in bad weather conditionst The structure andworking principle of ?thh m digital laser radar system is describedt ? MHzequivalent sampling frequency is realized by dual-channel parallel sampling tosample laser pulse echo signalt Multi-pulse coherent average algorithm isadopted to increase the SNR of pulse echo signal by N timest Time-of-flight isestimated by correlation detection method and the range resolution is furtherimproved to ?t?h m in theory by multi-time delayed correlating methodt Indoorexperimental results of rangin