Indoor GPS Technology Frank van Diggelen and Charles Abraham Global Locate, Inc.

Transcription

1 Indoor GPS Technology Indoor GPS Technology Frank van Diggelen and Charles Abraham Global Locate, Inc. Abstract It is well known that GPS, when used outdoors, meets all the location requirements or E911 as well as commercial location based services. The problem, till now, has been making GPS work indoors. This paper addresses the technical issues conronting indoor GPS, and shows, in clear and simple terms, how GPS technology can, and has been brought inside. We begin with a theoretical overview o the technology, and then show how the theory has been reduced to practice in a single baseband chip that perorms massively parallel correlations to detect the GPS signal at power levels 30dB (one thousand times) below those ound outside. The paper also addresses the role o the wireless operator in providing aiding inormation to the GPS receiver, the industry standards that have emerged, and the issues aced by anyone evaluating this new technology. Finally, the paper presents real world results showing the perormance o Global Locate s Indoor GPS hardware implementation. The layout o this paper is: 1. Introduction We give a preview o the beneits that accrue rom a new approach to high-sensitivity, wireless-aided, GPS. 2. Theory We describe the classical GPS approach, o separate acquisition and tracking unctions. We show why acquisition has traditionally been a slow process. This leads to the undamental idea o aiding/assisted-gps. We explain aiding and show the beneits to speed and sensitivity. We will see that the increase in sensitivity, just rom aiding, is limited to 10dB, and that or indoor operation one needs more than 20dB. This leads to the need or high-sensitivity receivers. We discuss the theory o high sensitivity receivers, showing that they must perorm a ull convolution, and integrate the results. We discuss an existing DSP-based implementation o a high-sensitivity receiver, and introduce the new hardware processing approach, which is the primary subject o this paper. 3. Worldwide network implementation Describes the worldwide reerence network constructed to provide GPS assistance to support indoor GPS. We also discuss the server that supports the position computation unction. 4. Indoor GPS hardware processing approach This section describes the new hardware technology developed to provide the 30dB gain over conventional GPS receivers, and thus provide indoor operation. 5. Field test results and summary In the inal section we show results o the combination o the Worldwide Network and the Indoor GPS chip, with positions calculated to great accuracy, outdoors, in urban canyons, and indoors. Finally we summarize the theory and new technology that has been discussed, to justiy the beneits that were claimed in the Introduction. Presented at CTIA Wireless-Agenda, Dallas, May Page 1 o 10 vd

2 Theory 1. Introduction This paper presents a new approach to highsensitivity, wireless-aided, GPS. The new design implements the correlation and integration unctions entirely in hardware, with a real-time convolution processor; an on chip integrator; and a ull-range loop that obsoletes the traditional separation o acquisition and tracking. We also introduce a worldwide network o tracking stations, that provide the aiding data used by the GPS hardware, and a server that perorms the position computation unction. The beneits o the new approach are: a. High sensitivity, even in environments with signiicant signal ading (i.e. indoors). b. No precision requency reerence required. c. No need or GPS time synchronization rom the wireless network. d. Ultra low CPU requirements with no DSP and no dedicated CPU. e. Autonomous or wireless-aided operation. The satellite transmits at a known requency o MHz, to which the satellite motion adds ±4.2 khz o Doppler shit. The speed o the GPS receiver adds 2.3 Hz/mph, and the uncertainty in the GPS receiver s local requency reerence adds an error o khz or each 1ppm o oscillator error. Thus there is a requency uncertainty o greater than ±4.2 khz on the observed GPS signal. The GPS receiver detects the signal by correlating. That is, multiplying the received signal with a locally generated replica o the code used in the satellite, and then integrating (or low-pass iltering) the product to obtain a peak correlation signal. The peak o this signal vanishes when the locally generated code-delay is wrong, or when the requency is wrong. Thus, to acquire the signal, a GPS receiver must search the entire space o possible requency osets and code delays, illustrated in Fig 2. The search is conducted over ranges o requency and code-delay, which we call bins. 2. Theory Traditionally GPS receivers have been designed with separate acquisition and tracking modes [1-3]. To compute a position, the device must irst acquire the satellite signals, and to do this it must search over all possible requency and code-delay bins. Why is this? Let s look at requency irst, and then code delay. Code Data MHz Code 1023 kbps Data 50 bps One bit o the code is called a chip The code is 1023 chips long, and repeats every millisecond Fig 1. GPS Signal at the Satellite Fig 2. Freq/Delay space, and correlation peak Classical GPS receivers have two to our correlators dedicated to each satellite. To detect the signal these correlators must be used with a locally generated code delay within one chip o the (unknown) correct delay. This implies a sequential search over 1023 possible chips, at each dierent requency bin. Typically there are Page 2 o 10

3 Theory 40 requency bins required to span the requency uncertainty. So the total search space is req/delay bins. Once the signal has been acquired, the receiver switches to tracking mode. But i it loses lock then acquisition must be repeated. Acquisition is a slow process. Typically GPS receivers have been designed to dwell or at least one millisecond in each req/delay bin, taking 40 seconds in all to search the entire req/delay space. This led to the undamental idea o aiding, or Assisted-GPS (A-GPS), proposed in 1981 [4]. The idea is to provide the receiver with inormation, such as the satellite ephemeris, rom which the receiver can estimate the satellite Doppler ahead o time, thereby dramatically reducing the required req/delay space that must be searched. Search in the box! Frequency bins rom aiding Fig 3. Freq/Delay search space, with aiding Fig. 3 illustrates the reduced search space that results rom aiding. The range o possible requencies can be reduced by a actor o ten, thus reducing the acquisition time by ten a signiicant improvement! Alternatively, with the right receiver architecture, instead o speeding acquisition, one could make use o the aiding to increase the dwell time in each bin, thereby increasing the sensitivity. Unortunately this does not increase sensitivity enough to allow indoor operation. Here s why [5]: i the search space is reduced by a actor o 10, then the dwell time in each bin could be increased rom 1 to 10 milliseconds, without changing the original total search time. Each extra millisecond o data can be integrated (summed) with the previous results, yielding SNR (signal to noise ratio) gains that approach N or each extra N milliseconds. Thus the aiding could produce a sensitivity gain that approaches 20log10 ( 10) = 10dB. Unortunately, GPS signal levels indoors are 20 to 30dB down rom the signal levels outdoors, and so aiding alone is not enough to make a receiver work indoors. High-sensitivity receivers work by perorming ar more correlations and integrations than a standard receiver in the same amount o time. With enough correlators, all possible delays (i.e. an entire convolution) can be calculated at the same time, and hundreds o milliseconds o convolutions can be integrated to give the required sensitivity or indoor operation. There are two existing methods o implementing an entire convolution, a DSP approach and a hardware processing approach. The DSP approach is a store-and-process technique that perorms the convolution in the requency (transorm) domain. Because it is a non realtime process, it cannot be used as part o a eedback loop, and so the DSP method relies on a-priori knowledge o the local oscillator requency. A hardware approach that matches the sensitivity o the DSP technique requires in excess o 8000 correlators [6]. Until recently this was not easible because the required chip would be too large (and expensive). Now, with 0.18 micron technology, Global Locate is producing a chip that has in excess o correlators. The packaged chip is 8mm 8mm. Details o this chip are discussed in Section 4. Page 3 o 10

4 Worldwide network implementation 3. Worldwide network implementation The GPS satellites transmit data that can be decoded by any receiver that has a clear line o sight to the satellite. There are, currently, 28 satellites in the GPS constellation. Thus, one only has to see all 28 satellites simultaneously to have access to all the data in real time. The most eicient and cost eective way o doing this is with a worldwide network o GPS reerence stations that eed data to a server. Such a network, once constructed, can support any number o A-GPS devices, anywhere. Hub SMLC PCF Location client GMLC MSC BSC MS Global Locate has designed and implemented this network. The network provides the aiding inormation required or A-GPS and, i necessary, the server can process the GPS measurements made at the phone. The network and server are novel in three ways: 1. The network is ully redundant, with stations placed round the world such that each GPS satellite can be seen at all times by at least two dierent stations. 2. The server includes a worldwide terrain model that gives the altitude o the surace o the earth, relative to the GPS datum. This allows the server to compute a position with ewer satellite measurements. Altitude aiding or A-GPS has been done using the altitude o nearby cell-towers but this leads to position errors when the phone s altitude diers rom the cell tower s by an unknown amount. Thus the worldwide terrain model improves accuracy, particularly in hilly terrain. The model comprises approximately one billion discrete grid points, with altitude known to 18-meter accuracy. 3. The server can compute position rom GPS pseudorange measurements, rom any device (by any manuacturer), without accurate GPS time tags. This means it can be deployed on networks such as GSM, W-CDMA, and US-TDMA, which are not synchronized to GPS time. CBC Location Client = requestor o location inormation GMLC = Gateway Mobile Location Center (applications, billing, maps, etc) SMLC = Serving Mobile Location Center (position, network interacing, etc) PCF = Position Computation Function MSC = Mobile Switching Center BSC = Base Station Controller CBC = Cell Broadcast Center BTS = Base Transceiver Station MS = Mobile Station (i.e. phone) Fig 4. Worldwide network and 3GPP implementation The Global Locate server has been implemented according to location services standards being developed by dierent standards bodies, such a 3GPP or GSM/UMTS, TR45 or US TDMA and 3GPP2 or CDMA/CDMA2000. Fig. 4 shows how the Global Locate server its into the proposed architecture or GSM [7]; it shows an SMLC and the corresponding logical network elements. The Global Locate server perorms the A-GPS PCF (Position Computation Function) o the SMLC as well as the assistance data management. Location services standards speciy two modes o position computation, MS Assisted, where the network computes the handset s position, and MS Based, where the handset computes position. The Global Locate server PCF supports both o these alternatives. In MS Assisted mode, the MS sends GPS measurements (pseudoranges) to the SMLC, BTS Page 4 o 10

5 Worldwide network implementation which computes positions. In MS based mode, GPS measurements collected at the MS are processed at the MS to yield a position ix. Satellite orbit inormation at the MS can be obtained rom the assistance data or, in avorable signal conditions (i.e. outdoors), directly rom the GPS satellites. According to the standards, assistance data can be delivered using either broadcast mechanisms or point-to-point. The Global Locate server supports both ormats. Broadcast assistance data is delivered using Cell Broadcast Messages via a Cell Broadcast Center. The broadcast mechanism is eicient since the satellite inormation contained is valid or a large geographic area and thereore it is exactly the same or many users. As a rule o thumb, the elevation o a GPS satellite (vertical angle above the horizon) changes by 1 degree every 100 kilometers, so users in the same general region see the same satellites above the horizon. Where the network inrastructure to support the standards is not yet in place, the server has been implemented using pre-standard interaces. Example: in a deployment in Europe the Global Locate network provides GPS assistance data to existing cellular/gps phones using SMS messages to deliver the data. Timing Precise timing is a signiicant implementation issue or indoor GPS. The satellite measurements need a time-tag so that the position computation unction can compute the location o the satellites at the time o transmission. The satellite ranges change at a rate o up to 800m/s, so a 1 second time-tag error would lead to an 800 meter range error, and, in turn, a position error o hundreds o meters. For this reason, time-tags o better than 10 millisecond accuracy have typically been a requirement or a position computation unction 1. Traditional GPS receivers get measurement time-tags directly rom the satellites, by demodulating time o week bits contained in the headers o the transmitted satellite data. Receivers indoors cannot demodulate the satellite data. Even high-sensitivity receivers, which can measure the code delay indoors, cannot demodulate the data. Thus indoor GPS receivers must get this timing inormation elsewhere, and they can only get it rom the cellular network i the network is synchronized to GPS time. US-CDMA networks are synchronized to GPS time. Other cellular networks are not. LMUs (Location Measurement Units) have been proposed to add GPS synchronization to networks to support A-GPS, but this implies a signiicant inrastructure build out. However, with the Global Locate PCF it is possible to provide a ull GPS assistance service (even or indoor GPS) without deploying a single LMU. This is because the Global Locate PCF employs a position computation algorithm that processes GPS pseudorange measurements to produce position without needing an accurate GPS time tag. This dramatically reduces the burden on an operator who must implement A-GPS support, since the Global Locate implementation requires no deployment o new inrastructure. 1 Readers amiliar with GPS algorithms will be aware that all GPS receivers must deal with a common-mode error that aects all pseudorange measurements by the same amount. This error is the result o all common propagation delays (cables, ilters, etc.) and any error in the oset o the locally generated code that is being used to measure each pseudorange. The common-mode error is oten, conusingly, reerred to as a clock-error. However, the common-mode error is not the same as a time-tag error. Page 5 o 10

6 Indoor GPS hardware processing approach 4. Indoor GPS hardware processing approach The new approach to high-sensitivity (indoor) GPS is based on real-time convolution o GPS signals over the entire range o possible code delays. A standard GPS receiver, with an early-late pair o correlators per satellite, can observe just one possible code-delay chip at a time. This is illustrated by the shaded bar in Fig 5. As discussed earlier (see Section 2) this receiver must search to acquire the signal beore it can track it. Fig 6 illustrates the new design, which makes use o a real-time convolution processor instead o an early-late correlator. The convolution processor contains over 2000 correlators per satellite, and can compute all possible correlation delays (i.e. a complete convolution) in real time. This design obsoletes the need or separate acquisition and tracking stages, since, no matter what the actual code-delay is, the output rom the convolution processor will always contain the correlation peak. In outdoor situations this means that signal acquisition occurs almost instantaneously. Fading signals In indoor environments there is requently signiicant signal ading. Even i a conventional GPS receiver manages to track the satellite at the low signal strength (usually by increasing the loop time constants), ading will cause requent loss o lock and return to the acquisition stage, where integration times are again limited by correlator resources. By contrast, the new design, which includes all possible code delays in the loop, can integrate continually, even when the signal is ading. With a large amount o integration (e.g. 1000ms) the SNR increases signiicantly over a receiver with 1ms o integration, enabling detection o approximately 23dB lower signal strengths. This gives us the sensitivity or indoor operation with signal strengths o 150dBm. Signal integration in hardware In standard GPS receivers accessing the correlators every millisecond generates signiicant CPU load. By including dedicated hardware or long term integration (up to several seconds) the new approach minimizes CPU interaction. There is no need or a dedicated CPU, as required in a conventional GPS design. The new chip can be integrated in a phone, and share the phone s CPU. The demands on the CPU are very low. Furthermore, there are no hard real-time constraints (interrupts are optional), and the GPS unction operates without interrupting voice calls. Acquisition Tracking IF Local Oscillator + e Early- Late Correlator integration ~1 ms Σ Frequency discriminator Search one code delay chip at a time IF hardware unctions Local Oscillator + e Real-time convolution processor Frequency discriminator integration ~1000ms Σ Search all code delay chips at a time aiding Fig 5. Standard GPS receiver Fig 6. Indoor GPS, hardware processing approach Page 6 o 10

7 Indoor GPS hardware processing approach DSP-based high sensitivity GPS Another approach to high-sensitivity GPS is a store-and-process DSP approach that perorms the convolution in the requency (transorm) domain. This is illustrated in Fig 7. The DSP approach makes use o the act that a time domain convolution is a simple multiply in the requency (transorm) domain, ater a Fourier Transorm. The technique works by irst storing a block o IF data in RAM (typically one second s worth), then perorming FFTs to yield the complete convolution in the time domain, which can be integrated to give high sensitivity. We will now contrast the new hardware processing approach with this DSP approach. No precise requency requirement The DSP approach is not a real time approach, and does not support requency adjustment rom the GPS signals. The DSP processing instead relies on having a very precisely known requency reerence, so that the local oscillator can be calibrated without eedback. This implementation is appropriate in certain wireless networks with very stable requency reerences (e.g. US-CDMA). e rom locked wireless signal Local Oscillator IF + e 2MHz aiding RAM The new hardware processing approach, because it is a real-time implementation, supports GPS-based requency adjustment. The eedback loop rom the convolution processor drives the local oscillator to produce the correct requency () needed to demodulate the IF signal. Thus, even with error ( e ) in the local oscillator, the action o the loop causes the receiver to remain at the correct requency. Thus this implementation is appropriate regardless o the wireless network in which the design is implemented. Autonomous or aided operation The DSP approach, as described in the previous paragraph, requires wireless aiding or the requency reerence. The new hardware approach can make use o aiding, such as satellite orbit data, to compute the satellite Doppler requency, and thereby assist the requency adjustment, by limiting the unknown requency to that caused by the receiver s velocity, and the local oscillator error. However, in a situation where aiding is not available, the hardware approach can still operate, by using the requency eedback and by demodulating the satellite data. In this mode the new design operates almost like a conventional GPS receiver, except that acquisition times are about one thousand times aster, thanks to the presence o the real time convolution processor. FFT IFFT integration 1000ms Σ Time Domain Z = X ( jω ) Y ( jω) = FFT( x( t) * y( t)) Z X Y FFT(PRN Code) Frequency (Transorm) Domain Fig 7. Indoor GPS receiver, DSP approach Page 7 o 10

8 Field Test Results 5. Field test results and summary The Global Locate hardware design or indoor GPS has been tested in many challenging environments where conventional GPS receivers do not work. Fig 9 shows a test in the dense urban canyon o the San Francisco inancial district. Fig 9a. Urban Canyon, San Francisco Fig 8a. Reinorced concrete parking garage Fig 8b. GPS Position Fig 8 shows a test inside a parking garage, two loors below the roo. Outside the structure is an apartment complex, blocking most open spaces in the walls. The Global Locate receiver computed positions to 20-meter accuracy, tracking and using 7 satellites. As a benchmark, standard GPS receivers tracked no satellites in the same test. Fig 9b. GPS Position In this test the Global Locate receiver calculated positions to 25- meter accuracy, tracking 7 satellites. Standard GPS receivers tracked up to two satellites in the same test, and could not compute a position. Page 8 o 10

9 Field Test Results In tests perormed in the Global Locate oices we leave the receiver running or many hours, generating scatter plots that show the perormance o the indoor GPS design. Next the antenna was placed inside a closed metal drawer, beneath two other metal drawers. Standard GPS receivers tracked no satellites in the same test; in act, the standard receivers could not compute a position when placed on top o the desk in the picture. Antenna Antenna inside Drawer open to show antenna. All data was collected with drawer closed Fig 10a. Inside 2-story oice building 30m circle Fig 10b. GPS Position In this test the receiver computed positions with a mean accuracy o 21 meters. Standard GPS receivers in the same test tracked between zero and two satellites, and could not compute position. The Global Locate receiver tracked and used up to 11 satellites. Fig 11a. Inside a closed metal drawer 30m circle In this test the Global Locate GPS receiver computed positions with a mean accuracy o 24 meters. The receiver tracked and used up to 10 satellites while inside the closed metal drawer. Fig 11b. GPS Position Page 9 o 10

10 Summary Summary We have described a new approach to indoor GPS, using Assisted-GPS with aiding rom a worldwide network o reerence stations, and with a hardware processing approach that includes a real-time convolution processor o over dedicated correlators. The theory, comparison with other architectures, and ield tests show the beneits o this design, namely: a. High sensitivity, even in environments with signiicant signal ading (i.e. indoors). See Sections 4 and 5. b. No precision requency reerence required. See Section 4. c. No need or GPS time synchronization rom the wireless network. See Section 3. d. Ultra low CPU requirements with no DSP and no dedicated CPU. See Section 4. e. Autonomous or wireless-aided operation. See Section 4. Reerences [1] GPS Signals, Measurements and Perormance, P. Misra & P. Enge, to be published Summer Contact Navtech Seminars and GPS Supply. [2] Understanding GPS, Principles and Applications, E. Kaplan, Editor, Artech House Publishers. [3] Global Positioning System: Theory and Applications, Vol I, Parkinson, Spilker, Axelrad & Enge, Editors, American Institute o Aeronautics and Astronatutics. [4] Navigation System and Method, United States Patent , Ralph Taylor and James Sennott, NASA, Filed May 22, [5] Indoor GPS, Wireless Aiding and Low SNR Detection, Course Notes: Navtech Seminars, Course 218, [6] An Introduction to Snaptrack Server-Aided GPS, Moeglein & Krasner. Proceedings o the Institute o Navigation conerence, ION-GPS [7] GSM Technical speciication Fig 12. Global Locate Indoor GPS chip Page 10 o 10