Multipath Error Reduction in Signal Processing Alexey Zhdanov, Victor Veitsel, Mark Zhodzishsky and Javad Ashjaee Javad Positioning Systems compensate for the imperfections of the above mentioned strobe methods. It should be noted here that such progress can be achieved without making any complicated changes to the receiver design or increasing its power consumption. BIOGRAPHY Alexey Zhdanov is with the JPS research and development group specializing in GNSS and signal processing. He has a Ph.D degree in EE. Victor Veitsel received his Ph.D. and Dr.Sc. in 1955 and 1967, respectively. Pr. Veitsel is presently Consultant at the Javad Positioning Systems. INTRODUCTION In practice, navigation signal processing comprises accumulation of resulting multiplication products of the received signal with both the reference carrier and reference code generated in the receiver. The reference carrier signal corresponds to the carrier for the given satellite and the reference code corresponds to the pseudorandom code of the same satellite. Devices that perform multiplication and accumulation of the products are called correlators, whereas the corresponding process is called correlation of both signals. It is known a lot of ways to design tracking channels in such a receiver, but the most commonly used design comprises three or four correlators. The first two correlators produce in-phase I and quadra-phase Q correlation signals. These I and Q signals are used to perform carrier synchronizing. For example, the phase lock loop (PLL) discriminator can function according to the algorithm Z d = arctan(Q / I ) . Signal I is used to demodulate information symbols as well. In different synchronization systems signal I represents the normalization signal. In both correlators the reference code is a replica of the satellite PRN code (Figure 1,a). Such a reference signal provides the closest to minimal both variance of the PLL tracking errors and probability of false demodulation of information symbols. The third (and fourth) correlator computes the inphase dI (quadra- phase dQ) signal, which is used in the delay lock loop (DLL) system. A discriminator of the coherent DLL can operate by the algorithm Z d = dI / I , while a discriminator of the noncoherent DLL functions with the following algorithm Z d = dI ⋅ I + dQ ⋅ Q . For both correlators the reference code is a strobe sequence, and hence the correlators are called strobe correlators Ref.[4]. Mark Zhodzishsky received his Ph.D. and Dr.Sc. in 1965 and 1984, respectively. Pr. Zhodzishsky is presently Research Group Manager at the Javad Positioning Systems. Javad Ashjaee pioneered high Precision GPS at Trimble Navigation. Dr. Ashjaee is founder of Ashtech (1987). Later he founded the JPS company, where he is closely involved in all aspects of product development. ABSTRACT The multipath error reduction methods described in this paper will give you an insight into further development of the so-called "strobe methods" that have been used in pseudo random noise (PRN) receivers since 1975 (see Ref.[1]). These methods already proved their high effectiveness in reducing tracking errors in spreadspectrum (PRN type) digital receivers (see Ref.[2-4]). However, it should be stressed here that they have some failings too, for example, the possibility for a DLL to lock on reflected signals, or problems due to very narrow DLL lock range. Hopefully we can now offer new techniques allowing engineers to reduce the failings inherent to the earlier strobe methods (these new techniques are patent pending). The idea of using strobes is based on replacing the 'regular' PRN-code replica in the receiver correlator with an alternative reference code replica built from a special series of strobes. In this paper, we will present a new effective strobe sequence and provide the reader with all the necessary theoretical information and experimental data on the expected noise and multipath errors. The new strobe sequence described below allows developers to PAGE To reduce the impact of multipath error due to varying strobe waveforms it has been suggested in Ref. [4] to generate relation dI (ε) similar to shown in Figure 2, b. It should be noted that the strobe, which we get in this case, is called the complex strobe. Such a shape of relation dI (ε) provides decreasing both the maximum peak of multipath error and the reflected signal influence zone. The maximum peak of multipath error ∆τmax is determined by the ratio of a value of maximum side peak dImax (see Figure 2) which is located on the right from the DLL operating point ε0 to the derivative Kd of dI (ε) with t a) ∆ D t b) D t c) D1 d) D2 t S1 A S2 respect to shift ε at the operating point ∆τ max = ±α dI max / K d , Figure 1. Reference signals where α is a ratio of amplitudes for reflected and line of sight signals. The reflected signal influence zone size is determined delay range where the multipath error for reflected signals with preset amplitude exceeds some threshold value. In Figure 2,b by convention this zone corresponds to the intervals (ε0 ÷ ε4) and (ε5 ÷ ε6 )]. EMBEDEMBEDFor example, a simple strobe (Figure 1, b) with duration D that is equal to front duration (transient process duration) of PRN-code modulating the incoming correlator signal provides almost minimal variance of noise errors in DLL tracking, i.e., this strobe is close to optimal. Note that DLL with simple strobe with duration D is equivalent early-late DLL with spacing D. It may be shown that in the quasilinear approximation the DLL errors are governed only by statistic of dI. The relation of the mean signal dI on the time shift ε between a reference and an input PRN code of this strobe is shown in Figure 2, a. Once being appeared the reflected signal results in generating a scaled replica of relation dI (ε) which is shifted to the left since the reflected signal delays always with respect to the line of sight signal, and as a rule, the reflected signal is weaker than the line of sight signal. As a result of adding these two relations dI (ε) , zero of the sum relation dI (ε) is offset. It causes offset of the discriminator characteristic (relation of the mean error signal Z d on the time shift ε) DISADVANTAGES OF THE DESCRIBED APPROACH AND WAYS TO THEIR AMENDMENT The approach described in INTRODUCTION has two important drawbacks • narrow DLL lock range; • possibility of DLL erroneous lock over reflected signals; Let us consider these disadvantages one after another. The narrow DLL lock range. We mean here the range of initial errors ε, at which the signals are synchronized in the operating point ε0 that has been selected. In Figure 2,b this range is equal to (ε1 ÷ ε2). This range is narrower than that for the DLL with a simple strobe (in Figure 2,a that range is equal to ±∆). If under signal energy search the adjusting delay step is large enough, for instance, it is ∆, then for correct acquisition a two-step acquisition procedure has been suggested in Ref.[4] to be used. After searching had been completed, we should turn on the DLL with a simple strobe at the first step of the acquisition procedure. After transient process in DLL had been completed, we should turn on a complex strobe at the second step of the acquisition procedure. Using such a two-step procedure increases acquisition time and requires extra resources to realize this simple strobe DLL. Note, that at the first sight we can replace the described two-step acquisition procedure with an extrasearch procedure, which is carried out after initial rough searching and has much less delay step. However, such an approach is inexpedient because there exists the inadmissibly great possibility to make an incorrect decision about position of signal’s maximum peak at extra-search, especially, in respect to the satellites with low carrier-to-noise power density ratios C/N0. If such a decision is made there will be an abnormally large error in pseudorange since in this case the DLL operating point Z d (ε) operating point, that results in generating the multipath error. dI(ε) b) ∆ ∆ ε0 ε3 ε 0 ε1 ε2 ε4 dImax ε5 (1) ε6 a) Figure 2. Possible shapes of relation dI (ε) PAGE will not coincide with the desired point ε0, but will be at either zero point with negative derivative (for example, at the point ε3 in Figure 2,b). Possibility of DLL erroneous lock over reflected signals. Such an erroneous lock may occur when there is a powerful enough reflected signal at the receiver input. This reflected signal is delayed by δ related to the line of sight signal over range (ε2 ÷ ∆) and its carrier phase θ with respect to carrier phase of the line of sight signal being close to zero. If C/N0 of the given satellite is low there exists a possibility of occurring a slip having approximately δ duration in DLL. As a result, DLL starts tracking the reflected signal instead of the line of sight one. To get rid of the drawbacks above we propose to synthesize a strobe providing relation dI (ε) of type in strobe locking type and a strobe shape. Remind that according to Ref.[4] two locking types are possible. They are: • locking with a moment of PRN code sign change, • locking with a PRN code chip edge. It should be noted that the strobe sequence of Figure 1, b has locking with a moment of PRN code sign change. Let us choose locking type first of all. It is easy to show that relation dI (ε) of Figure 3 may be only implemented using locking with a PRN code chip edge. Either strobe S (t ) having step waveform and total duration D can be generated from a sequence of n adjacent to each other rectangular pulses with magnitude Si (possibly zero) and delay di: n S (t ) = ∑ S i ⋅ (h0 (t − d i ) − h0 (t − d i +1 ) ) , Figure 3. If we choose ε0 as an operating point the maximum peak of multipath error will be determined by dImax [see Eq. (1)], while the reflected signal influence zone by range (ε0 ÷ ε4). where h0(t) is unit jump (Heaviside function), dn+1=D + d1. For example, if n=1 strobe S (t) is a rectangular pulse with duration D. Then the relation dI (ε) is equal to dI (ε) = dI(ε) ε0 ε3 ε ε2 ε4 0 k =0 n T i =1 0 dI (ε) = ∑ S i The acquisition range of DLL with such a relation dI (ε) is wide enough and equal to ∆, hence we need not employ the two-step acquisition procedure. If a delay step at the energy search stage is ∆, then after finishing this stage it is necessary to jump off by ∆/2 along shift ε to the left toward the flat area of relation dI (ε) . When transient ∑ µ k S (t − k∆) dt µ , (3) T / ∆ −1 µ k (h0 (t − d i − k∆) − ∫ C (t − ε) ∑ k =0 h0 (t − d i +1 − k∆) )dt (4) µ It is seen from Eq.(4) that for locking with a moment of PRN code sign change n d + d i +1 dI (ε) = ∑ S i ⋅dI S (ε − i , d i +1 − d i ) (5) 2 i =1 d + d i +1 , d i +1 − d i ) [in such a way where function dI S (ε − i 2 the multiplier in brackets ⋅ µ Eq.(4) is designed] process in DLL completes the operating point will be at ε0. While occurring the reflected signal distorts the left part of relation dI (ε) but it will not result in appearing extra zero points with negative derivative and hence in the DLL with such a relation dI (ε) erroneous lock over reflected signal will be impossible. PROVIDING T / ∆ −1 ∫ C (t − ε) If we use locking with a PRN code chip edge multiplier µ is equal ±1 dependent on the sign of PRN-code. In the case of locking with a moment of PRN code sign change if the sign of PRN-code changes, then µ will be equal to ±1 dependent on the sign. If the sign does not change µ=0. Substituting Eq.2 in Eq.3 and taking i-sum outside the integral symbol, we get dImax Figure 3. Desired shape of relation dI (ε) STROBE SYNTHESIS RELATION dI (ε ) T where C(t) is PRN-code passed through the radio frequency (RF) part and modulating the correlator incoming signal, T is accumulation time in the correlator, µ is a sign multiplier, ⋅ µ is ensemble µ average. ∆ 0 (2) i =1 represents nothing else than relation dI (ε) of a simple strobe (Figure 1, b) shown in Figure 2, a. This function has d + d i +1 and strobe two arguments; namely, shift ε − i 2 duration D = d i +1 − d i . So, relation dI (ε) of either strobe with step waveform can be generated as a linear combination of some number of simple strobe relation DESIRED To synthesize a strobe we will employ a heuristic approach. The synthesis will include both determination of PAGE • providing sufficient suppression of multipath errors; • providing a permissible level of energy losses (compared to the optimal strobe sequence). dI (ε) with multipliers Si. It is further evident, as function d i + d i +1 , d i +1 − d i ) (see Figure 2,a) is uneven 2 function, the definite integral of curve dI (ε) (see Eq.5) dI S (ε − 8 between limits -∝ and ∝ will be equal to zero regardless of selecting Si and di. But from Figure 3 it can be seen that the definite integral of desired curve dI (ε) does not have zero value. So, it is impossible to realize the desired relation dI (ε) using a strobe having locking with a moment of PRN code sign change. Therefore, we should employ locking with a PRN code chip edge. The simplest variant of the strobe sequence using locking with a PRN code chip edge is a strobe sequence shown in Figure 1, c. Such a sequence at duration D=73 ns (strobe with such duration is close to optimal) provides for GPS C/A code relation dI (ε) revealed in Figure 4, a. This 4 δ, m -8 0 0.04 a) B 0.00 b) ε, ns -0.04 400 200 300 400 So it appears to be expedient to set S1=S2 and D1=3D2. The relation dI (ε) for the case when S1=1 and D1+D2=146 ns is depicted in Figure 4, b calculated by 'Design Package' (see Ref.[5]). In Figure 5,b multipath error envelops at ratio of amplitudes α=0.5 for this strobe sequence is shown as well. For comparison in Figure 5,a there is multipath error envelopes for strobe showed in Figure 1,b. Level of synthesized strobe DLL energy losses compared to the simple strobe DLL (duration D=73 ns) is 5.1 dB. As a result of computation at DLL equivalent noise band BDLL=0.2 Hz and C/N0=40 dBHz the synthesized strobe DLL provides standard deviation of noise error about 0.53 m. In the case of GLONASS СA code, Figure 5 is valid except standard deviation. It will be 2 greater, i.e.; it will be equal to 0.76 m. 0.08 0 100 Figure 5. Multipath error envelops for strobe sequences showed in Figure 1, b and Figure 1, d dI(ε) -400 b) -4 0.12 -800 a) 0 relation dI (ε) will be provided if we use SAW-filter Sawtek 854668 with the bandwidth of 16 MHz at the level of -3 dB in RF-part. This described filter is employed in JPS receivers such as Legacy, Odyssey, Regency, and Eurocard. Figure 4, a shows that the amplitude span of the relation dI (ε) operating part (that area is designated as B) is very small and hence if noise is present signal tracking can be unstable. -1200 ∆τ(δ), m 800 EXPERIMENTAL DATA ON THE SYNTHESIZED STROBE DLL 1200 Figure 4 The relation dI (ε) for strobe sequences showed in Figure 1, c and Figure 1, d The easiest way to increase the amplitude span of the operating part is adding a negative pulse to the strobe (that pulse is designated as A in Figure 1, d). By varying duration D2 and amplitude S2 we can change both spacing and amplitude span of the operating part correspondingly. In the synthesis three key points have to be taken into consideration: • simplicity of implementing this strobe sequence; The synthesized strobe DLL has been tested in static and low kinematic scenarios. The main objective of testing was to find out to what extent the receiver performance can be improved by using the synthesized strobe DLL for the usual simple strobe DDL. Since multipath error reduction is especially important for differential tasks, we carried out our tests with the receivers running in the differential mode (with a few exceptions, see below). We connected two JPS Legacy (Ref.[6]) receivers to a common 'base antenna'. In the first Legacy receiver, the synthesized strobe DDL was enabled. The other Legacy was using the simple strobe DLL. The PAGE 'rover antenna' too was connected to (another) pair of Legacy receivers with similar strobe DLLs. Static Applications. In these tests, the base and rover antennas were installed on the roof of a high-rise. Since the two antennas were spaced several dozen meters aside there (and at different levels, too), the base and rover multipath errors could be considered uncorrelated. In fact, the roof had quite a complex surface with a number of signal reflecting details. Our preliminary analysis showed that the multipath delays 0 < δ <30 (meters) and relative amplitudes 0 < α < 0.5 were characteristic of that improvised test range. In our first experiment, all of the receivers were switched to the code differential mode providing smoothed GPS/GLONASS C/A pseudo ranges. These smoothed pseudo ranges were obtained with the help of a 50-second moving average filter where carrier phases were used for reducing code phase noise. Figures 6 and 7 show the differential positioning accuracy in plane and height, respectively. The black triangle indicates the estimates obtained for the receivers with the simple strobe DDL, whereas the gray bullet marks the data obtained with the use of the synthesized strobe DDL. All the relevant data were collected over a 14-hour observation interval. To get these data, we first had to compute a (precise enough) baseline estimate by using the carrier phase differential, which was then used as reference. Since the carrier phase multipath error has a zero mean (assuming a long enough interval), such a vector estimate can well be used as a 'truth' baseline. Figure 7. Height positioning errors in code differential As follows from Figures 6 and 7, the use of the synthesized strobe allows us to reduce the mean errors: in latitude, from 0.16 meters down to -0.02 meters, in longitude, from 0.52 meters to 0.06 meters, in height, from 0.52 meters to 0.04 meters. For standard deviation, this characteristic changed as follows: in latitude, from 0.89 m to 0.37 m, in longitude, from 0.52 m to 0.22 m, in height, from 1.38 m to 0.58 m. For maximal absolute value: in latitude, from 3.71 m to 1.83 m, in longitude, from 3.22 m to 1.31 m, in height, from 7.92 m to 2.51 m. Therefore, the use of the synthesized strobes results in considerable improving of receiver performance (we gain twice as much in terms of accuracy). In our second experiment, we were using the GPS/GLONASS carrier phase differential based on C/Asignal measurements. The main goal was to find out whether the code multipath reduction results in a smaller 'time to first fixed solution'. In the course of this 'cyclic' test, we repeatedly re-set the Real Time Kinematic (RTK) engine (Ref.[7]) as soon as the engine computed another fixed ambiguity solution. Both 'receiver pairs' (two receivers with the simple strobe DDL and another two with the synthesized strobe DDL), ran simultaneously for 70 hours with an average of 10 satellites being tracked during this time. Figure 8 shows the 'time to first fixed solution' histograms based on the collected data. The histogram shown in black corresponds to the receivers with the simple strobe DLL whereas the gray histogram describes the results relating to the synthesized strobe DDL. As seen from this figure, the use of the synthesized strobe DDL will also reduce the time it takes the receiver to fix biases and get a centimeter-level accuracy. In this experiment it allowed to reduce the mean time to first fixed solution of 8%. Figure 6. Plane positioning errors in code differential PAGE 0.20 2 'Code minus carrier', m 0.16 1 0.12 0 0.08 0.04 -1 time, s 0.00 time, s 0 10 20 30 -2 Figure 8. 'Time to first fixed solution' histograms 0 Low Kinematic Applications. Does a receiver with the synthesized strobe DDL make difference in low kinematic? This is a key question. In low kinematic, the receiver moves so slowly that the mechanism of "multipath randomization" is not fully enabled. Such scenarios are not uncommon: imagine a surveyor walking with his backpack kit in the field or a roller laying asphalt on the street etc. In this experiment, the rover antenna was mounted on the car moving at 4 mph. Again, two 'JPS Legacy' receivers were connected to this antenna. The next step to take might be to use a base antenna and another two receivers, but… Note that in tests like ours it is very important to assure very low multipath at the base antenna's location. Otherwise, it will be very difficult, if at all possible, to distinguish between the rover and base multipath. Obviously, this is a very tough requirement to meet. That is why we decided to make do without any base antenna/receivers in this test. Instead, we decided to use the so-called 'code minus carrier' combination (the difference between pseudo range samples and carrier phase samples expressed in meters) for the rover antenna. Figure 9 shows the code minus carrier combination based on the GPS PRN14 C/A-slot measurements from the receiver with the simple strobe (marked with the black triangle ) versus the analogous combination for the same satellite and slot but from the receiver with the synthesized strobe DLL (marked with the gray bullet ). Please note that for the sake of convenience the constant offset was removed from the combinations before drawing these plots. Figure 10 depicts similar data for another satellite (specifically, GPS PRN 25 C/A-slot). As follows from Figures 9 and 10 synthesized strobe DLL decreases values of code error jump caused be random reflectors two or more times. Therefore low kinematic applications, too, will benefit from using the synthesized strobe DDL since this results in a better signal tracking performance. 200 400 600 800 Figure 9. Code minus carrier for GPS PRN14 C/A 2 'Code minus carrier', m 1 0 -1 -2 time, s -3 1200 1400 1600 1800 2000 Figure 10. Code minus carrier for GPS PRN25 C/A CONCLUSION The use of the synthesized strobe DDL in GPS/GLONASS receivers results in a better signal tracking performance thus leading to higher positioning accuracies and availability. It resolves problems of the possibility for a DLL to lock on reflected signals and narrow DLL lock range. The conducted experiments showed the accuracy is improved twice as much in static code differential. The mean time to first fixed solution is reduced of 8%. The values of code error jump are decreased two or more times PAGE 4. Veitsel V., Zhdanov A., Zhodzishsky M. The mitigation of multipath errors by strobe correlators in GPS/GLONASS receivers. GPS Solutions, Volume 2, Number 2, Fall 1998. -pp. 38-45. 5. Zhodzishsky M., Vorobiev M., Zhdanov A and Ashjaee J. Automated Design of Navigation Receivers. Proc. of ION GPS-99, The Institute of Navigation, Nashville, Tennessee. 6. http://www.javad.com/Products/Hardware/ Legacy.htm. 7. http://www.javad.com/Products/Applications/ Real%20Time%20Kinematic%20Application.htm. in low kinematic applications. The gain is especially obvious in static code differential. REFERENCES 1. Zhodzishsky M. 1975. Digital PLL systems for video signal processing. Proc. of 2-nd scientific conference "PLL systems". Gorky, Russia (in Russian). 2. Garin L., Van Diggelen F. and Rousseau J.M. Strobe & Edge correlator multipath mitigation for code. Proc. of ION GPS-96, The Institute of Navigation, Alexandria, VA, 657-664. 3. Garin L. and Rousseau J.M. Enhanced strobe correlator multipath rejection for code & carrier. Proc. of ION GPS-97, The Institute of Navigation, Alexandria, VA, 559-568. PAGE