ENC GNSS 2006, Manchester, 7-10 May 2006

1

New Fast Signal Acquisition Unit for GPS/Galileo Receivers Wim De Wilde, Jean-Marie Sleewaegen, Andrew Simsky, Christophe Vandewiele, Edda Peeters, Jan Grauwen, Frank Boon Septentrio NV, Belgium [email protected]; [email protected]

1 Abstract The new GPS/Galileo signals have much longer code lengths than the GPS C/A code. This helps to mitigate inter-satellite cross-correlation and makes the signals more robust against interference. However, the dimensions of the search space increase with the code length. For the new codes, a traditional sequential acquisition algorithm is not practical because it results in unacceptable acquisition times. Parallel acquisition units that drastically improve acquisition performance have been presented in the past for military and indoor applications. These units are usually dedicated to the acquisition of one particular code. Because GNSS receivers of the next generation must be able to acquire signals with various code modulations, Septentrio developed a flexible hardware fast acquisition unit (FAU), which is capable of acquiring any of the new civil signals. This new acquisition unit uses a combination of a matched filter with an FFT to search through the range/Doppler space in a highly efficient manner. The uniqueness of this FAU is in its flexibility, which allows it to support all planned code modulations, including modernized GPS, Galileo and GSTB-v2, and GLONASS. This technology is used in the new Septentrio's GPS/Galileo receiver products. The reduction of the acquisition time significantly improves the performance of navigation GNSS receivers operating in highdynamic environment. In this paper the architecture and features of the new FAU are discussed. Several technical issues are highlighted, and the performance of the fast acquisition unit is discussed in comparison to the traditional approach.

2 Introduction GNSS satellites modulate an L-band RF carrier by a PRN code, which is in turn modulated by data symbols. At the start of the acquisition process, the incoming code phase and the Doppler frequency shift of a particular satellite are unknown to the receiver. The purpose of signal acquisition is to determine these parameters. This involves searching the code-phase/frequency search space for the signal. The area of the search space to a great extent defines the acquisition performance. This area is the

2

ENC GNSS 2006, Manchester, 7-10 May 2006

product of the code length times the frequency uncertainty; the latter is essentially the sum of the geometric Doppler uncertainty and the receiver clock error. Signal

GPS C/A GPS L2C

Centre Frequency (MHz) 1575.42 1227.6

GPS L5

1176.45

Galileo L1

1575.42

Galileo E5a

1207.14

Galileo E5b

1176.45

Current status

PRN (Chips)

length

Modulation type

Operational on all GPS satellites. Operational on 1 GPS satellite. Full constellation in 2009 Planned for WAAS in 2006. Full constellation in 2013 Operational on GIOVE-A. IOV constellation in 2008. Operational on GIOVE-A. IOV constellation in 2008. Operational on GIOVE-A. IOV constellation in 2008.

1023 20460

BPSK(1) BPSK(1)

10230

BPSK(10)

4092

BOC(1,1)

10230

BPSK(10)

10230

BPSK(10)

Table 1. Signal characteristics of some of the GPS and Galileo signals. As it can be seen in Table 1, a common characteristic of the codes of all the new GNSS signals (of modernized GPS as well as Galileo) is that they are much longer than the current GPS CA code. This implies that the search space is much larger for these new signals. If the future receivers shall use the same techniques for Galileo and modernized GPS as these currently in use for the GPS CA code, they will see a performance penalty in terms of acquisition speed at least by a factor of 10 (it will be demonstrated later that this is also true for the Galileo L1 code, which is only 4092 chips long). Summing up, the characteristics of the new signals force receiver designers to develop new fast acquisition techniques if they want to keep the level of acquisition performance adequate for real-life applications.

3 Acquisition process The acquisition involves correlating the incoming signal with a local signal replica, which is characterized with particular values of the frequency and the code phase. According to the traditional approach, the receiver sequentially scans all possible combinations of frequency/code-phase in the search space, until the correlation value exceeds a certain predefined threshold. Each pair of code-phase/frequency values defines a “cell”. Commonly used search granularities in the code phase and frequency dimensions are 0.5 code chips and 500Hz respectively, which determines the cell size. The time spent by the search process on each cell (the so-called “dwell-time”) determines the acquisition sensitivity of the receiver, i.e. the minimal power of the signal that the receiver will be able to find. It is a crucial design parameter of the acquisition algorithm: long dwell times improve sensitivity, but can rapidly lead to an unacceptable acquisition time due to the large number of cells to scan through. As an example, for a code-phase uncertainty of 10230-chips and a frequency uncertainty of +/-5kHz, the search space would contain more than 400000 cells if each cell has an area of 0.5chipsx500Hz.

ENC GNSS 2006, Manchester, 7-10 May 2006

3

The advantage of the sequential search is that it does not use any specific hardware resources: the acquisition re-uses the same correlators that are used later for the tracking of the signal. The number of correlators available for acquisition ranges from 2 in simple implementations to a dozen in more elaborate receivers. The number of correlators defines the number of cells that the sequential acquisition algorithm can explore simultaneously. Parallel search engines have been developed for high-sensitivity acquisition of the GPS-CA code, and for the direct acquisition of the P-code [1,2,3]. In these approaches, multiple correlation values are calculated within one integration interval on dedicated hardware, which has been optimized for the efficient calculation of correlation values. A standard technique is to compute the correlation in the frequency domain, using a circular convolution [6]. Another popular approach is to use a matched filter, often complemented with an FFT engine.

4 Septentrio’s parallel search architecture Currently existing parallel acquisition units are dedicated to a particular signal type (either to GPS-CA or to GPS-P codes). The introduction of a large variety of new signals, all of them with much longer codes than GPS-CA, requires the development of fast and fully flexible dedicated acquisition hardware, even for standard outdoor applications. The requirements for signal flexibility bring about new challenges for the receiver design: 1. The sampling rate of the acquisition unit must be flexible to accommodate different chipping rates. The sampling rate is the reciprocal of the search granularity in the code-phase dimension, and hence must be carefully adapted to the signal under consideration. 2. The acquisition unit must contain a flexible anti-aliasing filter prior to the sampling rate conversion in order to prevent correlation losses (related to item (1)). 3. The acquisition unit must support all foreseeable GPS, GLONASS, Galileo and GSTB-V2 modulations. Specifically, the unit must accommodate BOC(m,n), BPSK(n) and the AltBOC modulation types. 4. To mitigate potential interference in the new GPS/Galileo bands, the unit must support pulse blanking. 5. The PRN code characteristics must be fully flexible. In particular, the unit must be able to handle both primary and secondary codes when applicable. The Fast Acquisition Unit (FAU) developed by Septentrio meets all these challenges. Figure 1 represents a high-level diagram of the FAU.

4

ENC GNSS 2006, Manchester, 7-10 May 2006 Matched filter Baseband samples

I,Q, Anti-aliasing filter blanking

N-tap delay line M

I/Q Correlator

primary code

secondary code

Flexible Code generator

Code Buffer

Code rate

N

Coherent RAM, (filled column by column)

M-point FFT I2+Q2

Sample rate

rate generator

Max. peak detect

Non-coherent RAM

Figure 1. FAU principle. Using a combination of a N-tap matched filter and an M-tap FFT, the FAU scans a large area of cells in the code-phase/frequency uncertainty space in parallel. In this paper, the area covered by one run of the FAU is referred to as a “scan area”. All the cells in the scan area are scanned in parallel, which is equivalent to the use of several thousand correlators operating in parallel. The dimension of the scan area, and hence the amount of effective correlators can be flexibly selected depending on the hardware resources on a particular receiver model. The FAU performs successive coherent and non-coherent integration, and reports the position of the maximal correlation. The signal detection must be confirmed, and its precise position must be determined. This is done through a traditional sequential search algorithm, for example a Tong detector [4], which searches the signal in an interval of a few chips around the position detected by the FAU, and at a fixed Doppler as reported by the FAU. Since the search space is dramatically reduced via the FAU aiding, the search time of this sequential search is negligible. If the signal is confirmed, the tracking can start. If no, the FAU either proceeds to the next scan area or stops if all the scan areas have been searched.

ENC GNSS 2006, Manchester, 7-10 May 2006

5

Figure 2. Correlation values over the search space. An example of correlation amplitudes over the search space can be found in Figure 2. In this figure, the position of the signal is clearly visible. The FAU shall succeed in finding the signal if the signal correlation is larger than all the correlation peaks caused by noise. The coherent and non-coherent integration periods are the two parameters crucially important to the optimal FAU operation. These two integration periods must be carefully selected in order to achieve a given sensitivity. Because the number of noiseonly correlation peaks searched in parallel is huge, and the signal correlation must be larger than all of them, the integration periods must be larger than with a traditional sequential search algorithm. However, the high degree of parallelization in the FAU more than compensates for this drawback. In selecting the optimal durations of coherent and non-coherent integrations, the main criterion is the acquisition speed for a given sensitivity. After all, the main purpose of the FAU is to acquire the signal as fast as possible, without compromising the sensitivity. As will be shown below, improving the sensitivity is preferably achieved by increasing the non-coherent integration duration, rather than increasing the coherent integration.

5 Coherent versus non-coherent integration 5.1 Coherent integration The selection of the optimal duration of the coherent integration Tc is based on the compromise between two contradicting requirements: on the one hand, Tc must be reduced in order to increase the size of the scan area, which is inversely proportional to Tc in the frequency dimension. On the other hand, the longer Tc , the higher the sensitivity.

6

ENC GNSS 2006, Manchester, 7-10 May 2006

To gain 3dB in sensitivity, Tc must be made twice longer. This leads to an increase of the search time by a factor 4. Here a factor of two stems from the simple fact that the coherent integration duration is twice longer, and another factor of two is needed to compensate for the reduction of the size of the scan area by a half in the frequency dimension.

5.2 Non-coherent integration Another way to increase sensitivity is to rely on non-coherent integration, which implies integration of the power of the correlations computed over successive coherent integrations. Even if the SNR after the coherent integration is still much lower than 0 dB, it is possible to detect a peak by accumulating the correlations non-coherently during a sufficient time.

Number of non-coherent integrations

GPS-CA Galileo BOC(1,1) L5/E5a/E5b 2

10

1

10

0

10

-4

-2

0

2 4 6 post-coherent SNR [dB]

8

10

12

Figure 3.Number of non-coherent integrations required for a given sensitivity, given the 90% probability of detection. Figure 3 represents the number of non-coherent integrations needed to achieve the probability of detection of 90% assuming a given post-coherent SNR. It is apparent from Figure 3 that the number of required non-coherent integrations is higher with the Galileo E5a and similar signals than with the GPS CA code. The reason is that the code phase search space is 10 times larger for E5a than for GPS-CA, because the PRN code length is 10 times longer. For successful acquisition, the FAU must provide high enough amplitude of the signal correlation to supercede the noise correlation in all the cells in the search space. In order to increase the correlation value, a higher number of non-coherent integrations is required (that is more averaging).

7

ENC GNSS 2006, Manchester, 7-10 May 2006

It is often convenient to refer the post-coherent SNR value to the more common carrierto-noise values (C/N0) expressed in dB-Hz. This can be done by the following formula: C / N 0 [dB − Hz ] = SNR[dB ] − 10 * log 10(TC [s ]) ,

(1)

where Tc is the coherent integration time. A typical value of Tc is 1ms, in which case a post-coherent SNR of, say, 6dB corresponds to 36dB-Hz. To acquire the E5a signal at that level with a probability of detection of 90%, Figure 3 shows that 8 successive noncoherent integrations are needed. Figure 4 illustrates the corresponding probability of detection (Pd) as a function of the C/N0.

Probability of detection

1 0.8 0.6 0.4 0.2 0 20

25

30 35 40 C/N0 after correlation in the FAU [dB-Hz]

45

50

Figure 4.Probability of detection of the Galileo E5a signal (or the GPS L5 signal), with a coherent integration time Tc of 1ms and 30 non-coherent integrations. To gain 3dB, i.e. to be able to acquire signals at a SNR of 3dB (or 33dB-Hz if Tc = 1ms), Figure 3 learns that the non-coherent integration must be multiplied by a factor close to 3. As the slope of the curves in Figure 3 is essentially constant, this property is independent of the SNR (at least in the range of SNR shown in Figure 3 corresponding to typical receiver sensitivities). Because the total search duration is directly proportional to the number of non-coherent integrations, gaining 3dB therefore leads to a penalty of a factor 3 in acquisition time. Contrary to the coherent integration, increasing the number of non-coherent integrations has no detrimental effect on the size of the scan area.

5.3 Conclusion It has been shown in the previous two sections that gaining 3dB in sensitivity can be achieved by either doubling the coherent integration duration, or tripling the noncoherent integration time. It has also been shown that the total increase in the acquisition time is a factor of 4 in the first case, and a factor of 3 in the second case. This is a remarkable conclusion, as it appears that increasing the non-coherent integration is the most effective way of increasing the sensitivity. This is also good news as, in Galileo and modernized GPS,

8

ENC GNSS 2006, Manchester, 7-10 May 2006

the coherent integration duration is practically limited to the primary code length, which is 1ms in most cases. This conclusion only holds as long as the inverse of the coherent integration time does not exceed the frequency range of interest. Exceeding this would imply scanning meaningless frequency bins and wasting the FAU resources.

6 Galileo BOC(1,1) case The Galileo BOC(1,1) deserves special attention as this is the very first time that a BOC signal is available to civilian users. For the case of the Galileo BOC(1,1) signal on L1, two options have been proposed and discussed in the literature [5]: direct BOC acquisition or side-band acquisition. It will be shown that both methods are identical in terms of acquisition speed for a given sensitivity, and a given hardware complexity.

6.1

Direct BOC Acquisition

For a BPSK signal, the search granularity in the code-phase dimension is usually 0.5 chips. This leads to an average correlation loss of 1.15dB. Due to the multi-peak characteristics of the BOC(1,1) correlation, it is well known that the search granularity must be decreased in case of direct BOC acquisition [5]. However, it is often assumed that halving the granularity (i.e. using a spacing of 0.25 chips) is sufficient. This assumption is not true: in order to achieve the same average correlation loss as for a BPSK signal, the granularity must be divided by 3, i.e. it must be 0.16 chips in case of BOC(1,1). This is further clarified by Table 2. Condition BPSK, 0.5 chip spacing BOC(1,1), 0.5 chip spacing BOC(1,1), 0.25 chip spacing BOC(1,1), 0.16 chip spacing

Worst-case correlation loss [dB] 2.5 12 3.9 2.5

Average correlation loss [dB] 1.15 4 1.8 1.17

Table 2. Correlation losses caused by the search granularity. The reason why the search granularity must be divided by 3 is apparent when one compares the widths of the respective correlation peaks. It is shown in Figure 5 that the width of the main peak for BOC(1,1) is about one third of the width of the main peak of BPSK(1).

9

ENC GNSS 2006, Manchester, 7-10 May 2006

1

correlation amplitude

BOC(1,1) BPSK(1)

0.5

0

-0.5 -1.5

-1

-0.5

0 delay [chips]

0.5

1

1.5

Figure 5. Comparison of a BOC(1,1) and a BPSK(1) correlation peak. With the Galileo L1 code length of 4092 chips, the above discussion indicates that searching the BOC(1,1) Galileo L1 signal in direct acquisition mode requires scanning of 4092/0.16=24552 code phases. A BPSK modulation with the same power and code length would only require 4092/0.5=8184 code phases to scan. This leads to a penalty of a factor of 3 in the acquisition time with respect to a BPSK modulation with the same power.

6.2

Side-Band Acquisition

On the other hand, a side-band acquisition can use the same granularity as a BPSK (i.e. 0.5 chips), but must recover from the 3dB loss caused by the fact that the signal power in the side peak is a half of the signal power in the main peak. As already discussed, the most efficient way to recover 3dB loss is to multiply the non-coherent integration period by 3.

6.3

Conclusion

It can be concluded that side-band acquisition is equivalent to direct acquisition. Both suffer from a penalty of a factor of 3 in acquisition time with respect to an equivalent BPSK signal. The FAU presented in this paper supports both modes of operation. It is worth noting that the Galileo L1 code, though apparently the shortest of all Galileo codes (having only 4092 chips), has an effective length of 3*4092=12276 as far as the acquisition performance is concerned. In other words, in terms of acquisition time for a given sensitivity, acquiring the Galileo BOC(1,1) code is equivalent to acquiring a BPSK code of 12276 chips. This means that Galileo L1 code is the most difficult to acquire as compared to all the other Galileo codes. It will be faster for a receiver to acquire the E5a or E5b codes first, and then handover to L1.

10

ENC GNSS 2006, Manchester, 7-10 May 2006

7 Measurement results All the future Septentrio receivers shall be Galileo-enabled, and they shall all be equipped with the FAU unit. The FAU unit is already used in the latest Septentrio’s GeNeRx Galileo/GPS all-signals receiver. Using this receiver, a test has been run in order to compare the performance of a FAUbased signal search versus a traditional sequential search algorithm based on a Tong detector. In both cases, the search space is as follows: 1. Code-phase uncertainty: the whole code length (no a-priori code-phase information available); 2. Frequency uncertainty: +/-8kHz. The sequential search algorithm used 8 correlators, to scan 8 cells in parallel. The FAU contained the equivalent of 16384 correlators. Signal type

Acquisition sensitivity (90%) [dB-Hz] GPS-CA 40 GIOVE-A L1-BOC(1,1) (*) 40 GIOVE-A E5a, E5b 40 GIOVE-A E6-BPSK(5) (*) 40 (*) The GSTB-V2 PRN code for these signals is twice acquisition will be twice faster.

Search Time FAU [s]

0.06 1.67 0.48 0.48 longer as the one used in

Search Time Tong [s]

10 240 100 100 Galileo, hence the Galileo

Table 3. Measured acquisition times for different signals. The performance difference is very clear, with a factor of up to 200 in favor of the FAU. At first sight, one could expect a performance increase by a factor equal to the ratio between the number of correlators in both cases: 16384/8=2048. However, actual gain in the performance is smaller due to the two reasons as follows: 1. As already discussed, the presence of a large number of noise-only correlations in the scan area requires longer averaging time in order to ensure that the signal correlation is dominant over all the noise correlations. 2. Additional losses in the FAU hardware must also be compensated by longer averaging.

8 Conclusions All the signals of Galileo and the modernized GPS will use much longer codes than the current GPS CA code. This implies that traditional acquisition strategies based on the sequential search through the time/frequency search space will become impractical. To achieve short acquisition times that would satisfy the requirements of real-life highdynamic applications, dedicated fast acquisition units (FAU) must be used. The uniqueness of the FAU developed by Septentrio is in its ability to support all the signals of Galileo and future GPS in a wide range of operational conditions. This FAU has already been used to acquire all current GPS and GSTB-V2 signals. Presented

ENC GNSS 2006, Manchester, 7-10 May 2006

11

performance results show an advantage by 2 to 3 orders of magnitude compared to the traditional sequential search approach. The optimal search strategy for the new signals has been discussed. It has been shown that non-coherent integration is generally the optimal way to increase the sensitivity. A section has been dedicated to the Galileo L1 BOC(1,1) signal, which is the only BOC signal currently considered for civilian users. It has been demonstrated that the effect of the BOC(1,1) sub-modulation is a penalty of a factor of 3 in acquisition time compared with BPSK signals. This provides an additional indication that the dedicated fastacquisition units are absolutely needed for future Galileo/GPS receivers.

9 Acknowledgements This work has been partly funded by ESA-ESTEC Contract 17932/03/NL/DS. The authors are thankful to Martin Hollreiser and Marco Falcone of ESA for excellent followup of the FAU development in the framework of Galileo receiver projects.

10 Bibliography 1. Rounds, S., Norman, C. Combined Parallel and Sequential Detection for Improved GPS Acquisition. Proceedings of the IAIN, June 2000, San Diego, CA, USA. 2. Eerola, V. Rapid Parallel GPS Signal Acquisition. Proceedings of ION GPS 2000, September 2000, Salt Lake City, UT, USA. 3. Betz, J.W., Fite, J.D., Capozza, P.T. Getting to M. Direct Acquisition of the New Military Signal, GPS World, April 2005, pp. 40-46. 4. Ward, P. Satellite Signal Acquisition and Tracking, in “Understanding GPS Principle and Applications”, edited by E.D. Kaplan, Artech House, 1996. XX 5. Martin, N., Leblond, V., Guillotel, G.,Heiries, V. BOC(x,y)signal acquisition techniques and performances, Proceedings of ION GPS 2003, September 2003, Portland, OR, USA. 6. Van Nee, D.J.R, Coenen, A.J.R.M. New Fast GPS Code-Acquisition technique using FFT, Electronics Letters, Vol. 27, No. 2, 17th January 1991, pp. 158-160.