Joint GNSS and 3GPP-LTE based Positioning in ...

Joint GNSS and 3GPP-LTE based Positioning in Outdoor-to-Indoor Environments - Performance Evaluation and Verification Armin Dammann, Emanuel Staudinger, Stephan Sand, Christian Gentner German Aerospace Center (DLR) Institute of Communications and Navigation Oberpfaffenhofen, 82234 Wessling, Germany Email: {Armin.Dammann, Emanuel.Staudinger, Stephan.Sand, Christian.Gentner}@DLR.de B IOGRAPHY Armin Dammann studied electrical engineering at the University of Ulm, Germany, with main topic informationand microwave-technology. He received the Dipl.-Ing. and Dr.-Ing. (PhD) degree in 1997 and 2005 respectively, both from the University of Ulm. In 1997 Armin Dammann joined the Institute of Communications and Navigation of the German Aerospace Center (DLR). Since 2005 he is head of the Mobile Radio Transmission research Group. His research focus currently includes synchronization/positioning for terrestrial OFDM systems. Emanuel Staudinger studied Hardware/Software Systems Engineering at the University of Applied Sciences in Hagenberg, Austria, where he received his B.Sc. degree in 2008. He continued his studies at the same university and received his M.Sc. in Embedded Systems Desgin in 2010. He is currently working towards the Ph.D. degree at the Institute of Communications and Navigation of the German Aerospace Center (DLR), Germany. His current research interests include synchronization in OFDM-based receivers, positioning algorithms for 3GPP-LTE, impact of analog receiver imperfections for these receivers, physical layer design for cooperative positioning, as well as implementation and rapid prototyping of these algorithms on FPGAs. Stephan Sand received the M.Sc. degree in electrical engineering from the University of Massachusetts Dartmouth, USA and the Dipl.-Ing. degree in communications technology from the University of Ulm, Germany in 2001 and 2002. In 2010 he obtained the Dr. Sc. ETH Zurich in communications technology from the Swiss Federal Institute of Technology (ETH) Zurich, Switzerland. Since February 2010, Stephan has been certified by the Project Management Institute (PMI) as a Project Management Professional (PMP). He is currently managing and working on multi-sensor navigation research projects at the Institute of Communications and Navigation, German Aerospace Center (DLR), Oberpfaffenhofen, Germany. Stephan has authored and co-authored more than 60 technical and scientific publications in conferences and

journals in the areas of wireless communication and multisensor navigation. Christian Gentner studied electrical engineering at the University of Applied Science in Ravensburg, Germany, with the main topic communication technology. He received his Dipl.-Ing. (BA) degree in 2006. He continued his studies at the University of Ulm, Germany until 2009, where he received his M.Sc. in telecommunication and media technology. He is currently working towards the Ph.D. degree at the Institute of Communications and Navigation of the German Aerospace Center (DLR), Germany. His current research interests include multi-sensor navigation, propagation effects and non-line-of-sight identification and mitigation as well as the implementation of these algorithms on FPGAs. A BSTRACT Indoor positioning is an extremely challenging task for GNSS positioning. Thus, we propose to use terrestrial communications systems such as 3GPP-LTE as a complementary positioning system. We estimate the expected positioning performance of 3GPP-LTE indoor positioning by assessing link budgets and calculating the Cramér-Rao lower bounds for pseudo-range and 2D position estimation. An experiment shows that the combination of mobile radio based positioning can achieve a positioning accuracy in the range of a few meters, indicating that such technologies are suitable for complementing GNSS indoors. I. I NTRODUCTION The position of mobile devices is important information in many areas of daily life. Applications like navigation or location based services benefit from accurate position determination. Even in the field of future wireless communications systems, the processing of position information has gained increasing interest and allows optimization of key figures like throughput or spectral efficiency. Global Navigation Satellite Systems (GNSS) like the US GPS or the European Galileo system provide position information with sufficient accuracy in many environments. In urban areas or even indoors, where

requirements on accuracy and availability are potentially high, GNSS based positioning suffers from non line-of-sight propagation (NLoS) or weak received signal power, resulting in insufficient positioning accuracy or availability. To overcome this problem, complementary positioning using signals from mobile communications systems has recently gained an increasing attention. In this paper we evaluate the performance of combining GNSS and mobile radio based positioning using the 3GPPLTE mobile radio communication standard, which is currently deployed in many countries. The downlink of 3GPPLTE is based on Orthogonal Frequency Division Multiplexing (OFDM), which allows a spectral efficient and flexible usage of the available frequency spectrum. The 3GPP-LTE standard specifies signal parts, dedicated to time and frequency synchronization. Both, the so-called Primary Synchronization Signal (PSS) and the Secondary Synchronization Signal (SSS) is an OFDM symbol within the 3GPP-LTE downlink frame. For our investigations we’ll use these synchronization signals together with scattered pilots, which are originally dedicated to channel estimation, for signal propagation delay based 2D position estimation. We exemplarily consider an outdoor-toindoor scenario, where the transmitters of a wireless communications system — representing the 3GPP-LTE base stations — are located outside an office building and the mobile terminal moves along a corridor indoors. First, we evaluate the achievable performance by calculating the link budgets from the transmitters to the receiver and applying them to the Cramér-Rao lower bound for 2D positioning. We compare these theoretical results to those which we got from an experiment in this environment.

is a superposition of 𝑁 modulated subcarriers, usually centered at frequency zero. Sampling at time instances 𝑘 𝑇 = 𝑘 𝑇OFDM = 𝑘 𝑁 𝑓1SC , 𝑘 = 0, . . . 𝑁 − 1 yields 𝑁 1 𝑠(𝑘 𝑇 ) = √ 𝑁

1 𝑠(𝑡) = √ 𝑁

⌊ 𝑁 2−1 ⌋

∑ 𝑛=⌊− 𝑁 2−1 ⌋

𝑆𝑛 e𝑗 2𝜋 𝑛𝑓SC 𝑡 .

(1)

𝑆𝑛 e𝑗

2𝜋 𝑁

𝑛𝑘

𝑛=⌊− 𝑁 2−1 ⌋

𝑁 −1

𝑁 −1 ∑

𝑆𝑛−𝑁 e𝑗

2𝜋 𝑁

𝑛𝑘

,

𝑛=⌈ 𝑁 2−1 ⌉

(2) 2𝜋 𝑁

2𝜋 𝑁

where we have used e𝑗 (−𝑛) 𝑘 = e𝑗 (𝑁 −𝑛) 𝑘 . The second part of Eq. (2) shows that the time domain signal samples can be obtained using a Discrete Fourier Transform (DFT) together with an appropriate reordering of the data symbols 𝑆𝑛 . The normalization factor √1𝑁 ensures that the energies of the signal sample vectors are equal in both time and frequency domain. Neglecting noise, we obtain 𝑁 signal samples of the OFDM symbol at the receiver as 1 𝑠(𝑘 𝑇 − 𝜏 ) = √ 𝑁

⌊ 𝑁 2−1 ⌋

∑

𝑆𝑛 e𝑗 2𝜋 𝑛𝑓SC (𝑘 𝑇 −𝜏 )

(3)

𝑛=⌊− 𝑁 2−1 ⌋

with 𝑘 = 0, . . . , 𝑁 − 1. 𝜏 is the signal delay from the transmitter to the receiver, which we wish to estimate. B. Cramér-Rao Lower Bound for Timing and Pseudo Range Estimation in OFDM We assume uncorrelated complex valued Gaussian { noise } 2 samples 𝑛(𝑘) with zero mean and variance 𝜎 2 = E ∣𝑛(𝑘)∣ and consider the signal model

II. OFDM S IGNAL M ODEL AND P ERFORMANCE A S -

Today’s terrestrial broadband communications systems like IEEE 802.11 (WLAN), IEEE 802.16 (WiMAX) or 3GPPLTE are based on Orthogonal Frequency Division Multiplexing (OFDM) [1]. OFDM is a multi-carrier modulation scheme which provides high spectral efficiency and high flexibility in assigning time-frequency resources to users at low to moderate signal processing complexity at both transmitter and receiver. In OFDM systems, the complex baseband transmit signal is usually designed and described in frequency domain. An OFDM symbol consists of 𝑁 complex valued data symbols 𝑆𝑛 chosen from modulation alphabets like e.g. 4-QAM or 16QAM. These data symbols are modulated onto 𝑁 subcarriers. The subcarrier spacing 𝑓SC is chosen such that the subcarriers are orthogonal. Orthogonality holds for a subcarrier spacing 1 of 𝑓SC = 𝑇OFDM , where 𝑇OFDM is the duration of an OFDM symbol. The OFDM TX baseband signal in time domain

∑

⌊ 2 ⌋ 2𝜋 1 ∑ 1 𝑆𝑛 e𝑗 𝑁 𝑛 𝑘 + √ =√ 𝑁 𝑛=0 𝑁

𝑟𝑘 = 𝑠(𝑘 𝑇 − 𝜏 ) + 𝑛(𝑘),

SESSMENT

A. OFDM Signal Model

⌊ 𝑁 2−1 ⌋

𝑘 = 0, . . . , 𝑁 − 1.

(4)

Using this samples we wish to determine an estimate 𝜏ˆ for the delay parameter 𝜏 . In this case the Cramér-Rao Lower Bound is defined as [2] Var[ˆ 𝜏] ≥

𝜎2 2 . ∑𝑁 −1 d 2 𝑘=0 , d𝜏 𝑠(𝑘 𝑇 − 𝜏 )

(5)

Using the properties of the DFT we have ⌊ 𝑁 2−1 ⌋

𝑁 −1 ∑ 𝑘=0

2 d 2 = 4𝜋 2 𝑓SC 𝑠(𝑘 𝑇 − 𝜏 ) d𝜏

∑

2

𝑛2 ∣𝑆𝑛 ∣ . (6)

𝑛=⌊− 𝑁 2−1 ⌋

Inserting this result into Eq. (5) leads to 𝜎2

Var[ˆ 𝜏] ≥ 2 8𝜋 2 𝑓SC

⌊ 𝑁 2−1 ⌋ 𝑛=⌊− 𝑁 2−1 ⌋

∑

2

.

(7)

𝑛2 ∣𝑆𝑛 ∣

For the case of OFDM, this result shows the well known dependency of the Cramér-Rao lower bound on the spectral properties of the transmitted signal. Spending the available signal energy for the subcarriers at the spectrum edge maximizes the sum in the denominator or, equivalently, provides better timing estimation performance.

C. Link Budget Assessment

2

10

SNR = 0 dB SNR = 15dB SNR = 30dB SNR = 45dB SNR = 60dB

1

10

In our signal model introduced in Eq. (4) we have assumed uncorrelated white Gaussian noise samples. This is achieved for a noise bandwidth of

0

RMSE [m]

10

𝐵=

−1

10

1 = 𝑁 𝑓SC . 𝑇

(12)

With the Boltzman constant 𝑘B = 1.38 × 10−23 Ws K and a temperature of 𝑇 = 300 K we get the power of the noise signal as 𝑃N = 𝜎 2 = 𝑘B 𝑇 𝐵 = 101 dBm, (13)

−2

10

−3

10

−4

10

0

5

10 Bandwidth [MHz]

15

20

Fig. 1. Cramér-Rao lower bound for pseudo range estimation using OFDM signals.

Eq. (7) can be simplified if we assume an equal distribution of the signal energy { 2 ∣𝑆∣ , ∣𝑛∣ ≤ 𝑀, 2 ∣𝑆𝑛 ∣ = (8) 0, else among used subcarriers 𝑛 = −𝑀, . . . , +𝑀 , which are symmetrically grouped around subcarrier zero. Using Eq. (7) leads to Var[ˆ 𝜏] ≥

2 8𝜋 2 𝑓SC

𝜎2 2 ∑𝑀 2 ∣𝑆∣ 𝑛=−𝑀 𝑛 =

∑

(𝐿 𝑛)2 = 𝐿2

𝑛=−𝑀/𝐿

∣𝑆∣2 𝑀 (𝑀 +1)(2𝑀 +1) 𝜎2 3

𝑀/𝐿(𝑀/𝐿 + 1)(2𝑀/𝐿 + 1) 3

. (9)

(10)

3

2𝑀 , 3𝐿 assuming that 𝑀 can be divided by 𝐿 without remainder. Fig. 2 shows the Cramér-Rao lower bound for pseudo-range estimation in form of the root mean √ √ squared error RMSE = 𝑐0 Var[ˆ 𝜏 ] = 3 × 108 m/s × Var[ˆ 𝜏 ] versus the signal bandwidth 𝐵 = (2𝑀 + 1) 𝑓SC for 𝐿 = 16 and different subcarrier signal-to-noise ratios 𝑀/𝐿≫1

≈

∣𝑆∣2 . (11) 𝜎2 The strong dependency on the occupied bandwidth is clearly visible. Again, this indicates the value of high bandwidth signals for timing respectively ranging estimation. SNR =

SNR = 10 log

𝑁 ∣𝑆∣2 = 𝑃RX − 𝑃N + 10 log , 𝜎2 𝑁u SNR = 60.6 dB

We may generalize this result in case we use every 𝐿th subcarrier only but keeping the occupied bandwidth constant. In this case the sum in the denominator of Eq. (9) modifies to 𝑀/𝐿

where 𝑃TX is the transmitted signal power and 𝐺TX , 𝐺RX are the antenna gains at transmitter and receiver. We assume free space propagation, where the decay factor 𝛾 = 2. For a carrier frequency of 2.45 GHz (𝜆 = 0.12 m), antenna gains 𝐺TX = 𝐺RX = 0 dB and TX power 𝑃TX = 28 dBm, we get a received signal power of 𝑃RX = −52.4 dBm at a distance of 𝑑 = 100 m between transmitter and receiver. Since we observe this signal power in 𝑁u used subcarriers, we get as subcarrier SNR (15)

which results in 1

2 8𝜋 2 𝑓SC

for a system bandwidth of 𝐵 = 20 MHz. In free space, we obtain the received signal power in logarithmic scale as ( ) 4𝜋 𝑑 𝑃RX = 𝑃TX + 𝐺TX + 𝐺RX − 10 𝛾 log , (14) 𝜆

(16)

for the system parameters mentioned above. Note, 𝑁/𝑁u = 𝐿 = 16 in our case. D. Positioning Performance Assessment Subsequently, we calculate the Cramér-Rao lower bound for 2D positioning. We start with the pseudo-range error model √ 𝑑˜𝑖 = (𝑥 − 𝑥𝑖 )2 + (𝑦 − 𝑦𝑖 )2 + 𝑐0 Δ𝑇𝑖 + 𝜖𝑖 (17) for 𝑖 = 1, . . . 𝑁BS , where [𝑥, 𝑦] is the mobile terminal (MT) position and [𝑥𝑖 , 𝑦𝑖 ] denotes the location of base station (BS) 𝑖. 𝜖𝑖 describes the pseudo-range error. We assume 𝜖𝑖 to be mutually uncorrelated Gaussian distributed with zero mean and variance 𝜎𝑖2 . Δ𝑇𝑖 is the difference of the time bases between BS 𝑖 and the MT. In case of fully synchronized BSs Δ𝑇𝑖 = Δ𝑇 is equal for all BSs, so we have three unknown parameters [𝑥, 𝑦, Δ𝑇 ] to estimate. However, in our experimental setup (4 transmitters), we have pairwise synchronous BSs, where pair BS1 and BS2 have a synchronized time base with difference Δ𝑇1/2 to the MT. Accordingly, Δ𝑇3/4 is the time base difference of BS pair BS3 and BS4 to the MT. In this case the parameter set which we have to estimate is [𝑥, 𝑦, Δ𝑇1/2 , Δ𝑇3/4 ].

√ Using 𝑑𝑖 = (𝑥 − 𝑥𝑖 )2 + (𝑦 − 𝑦𝑖 )2 , i.e., the distance between MT and BS𝑖 , and the derivatives d ˜ 𝑥 − 𝑥𝑖 𝑑𝑖 = d𝑥 𝑑𝑖 𝑦 − 𝑦𝑖 d ˜ 𝑑𝑖 = d𝑦 𝑑 { 𝑖 𝑐0 , d 𝑑˜𝑖 = dΔ𝑇1/2 0, { 0, d 𝑑˜𝑖 = dΔ𝑇3/4 𝑐0 ,

BS2

BS1

(18) (19) 𝑖 = 1, 2 𝑖 = 3, 4

BS3

(20)

𝑖 = 1, 2 𝑖 = 3, 4

21 m

(21)

of the pseudo-range signal model (see Eq. (17)) with respect to the parameters to be estimated we can construct the Fisher Information Matrix (FIM) as I = G Σ−1 GT with

BS4

(22) 0

⎛ 𝑥−𝑥1 𝑑1 ⎜ 𝑦−𝑦1 ⎜ 𝑑1

G=⎜ ⎝ 𝑐0 0

𝑥−𝑥2 𝑑2 𝑦−𝑦2 𝑑2

𝑐0 0

𝑥−𝑥3 𝑑3 𝑦−𝑦3 𝑑3

0 𝑐0

𝑥−𝑥4 𝑑4 𝑦−𝑦4 ⎟ 𝑑4 ⎟

0.4

⎞

⎟ 0 ⎠ 𝑐0

(23)

0.8 1.2 RMSE [m]

outer wall

1.6

inner wall

2.0

core wall

Fig. 2. Cramér-Rao Lower Bound for 2D position estimation with 2 pairs (BS1 -BS2 and BS3 -BS4 ) of synchronous BS.

and [ ] Σ = diag 𝜎12 , 𝜎22 , 𝜎32 , 𝜎42 .

(24)

From Eqs. (9), (14) and (15) we observe that the pseudorange 𝜎𝑖2 are proportional to the propagation loss ( 4𝜋 𝑑 )variances 𝛾 𝑖 , which we assumed to be free space (𝛾 = 2). For 𝜆 our evaluations we also take into account that a radio signal from BS𝑖 to the MT may travel through walls, which cause additional signal power attenuation 𝐿𝑖 . 𝐿𝑖 itself depends on the material of which the walls consist and the number of walls which are located in the signal path. Thus we obtain ( )𝛾 𝑑𝑖 2 𝜎𝑖2 = 𝜎ref 𝐿𝑖 . (25) 𝑖 𝑑ref In Sec. II-C we have estimated the subcarrier SNR = 60.6 dB at a distance of 𝑑ref = 100 m.( For a signal) bandwidth of 2 2 −4 𝐵 = 20 MHz, we get 𝜎ref m from Fig. 2 𝑖 = 7 × 10 respectively Eq. (9). We obtain the Cramér-Rao lower bound CRLB2D = (I−1 )1,1 + (I−1 )2,2

(26)

for 2D positioning by summing up the first two diagonal elements of the inverse FIM. Fig. 2 shows a color plot of the Cramér-Rao lower bound for 2D positioning according to Eq. (26). We modeled the considered outdoor-to-indoor scenario using a simplified floor plan. Wall penetration losses are summarized in Table I. III. E XPERIMENTAL S YSTEM A. 3GPP-LTE Frame Structure Generally, there are two different types of 3GPP-LTE radio frames, applicable to frequency division duplex (”Type 1”) and time division duplex (”Type 2”). We are focussing on the

TABLE I WALL PENETRATION LOSSES Wall type outer wall inner wall core wall

Penetration loss 32 dB 4 dB 16 dB

FDD frame ”Type 1”. Such a radio frame is 10 ms long and consists of 10 subframes. Each of them are built from two socalled slots. So, each downlink slot has a length of 0.5 ms and consists of 3, 6 or 7 OFDM symbols, dependent on the cyclic prefix length mode and the subcarrier spacing 𝑓SC , i.e., the core OFDM symbol length 𝑇OFDM = 1/𝑓SC = 15 1kHz = 66.67 𝜇s. Note, there is an optional subcarrier spacing of 𝑓SC = 7.5 kHz, which reduces the guard interval overhead but is more sensitive against Doppler. Fig. 3 shows the frame structure for the 72 core subcarriers of the spectrum containing synchronization signals and control channels. The different physical channels are ∙ Primary Synchronization Signal (PSS, blue), ∙ Secondary Synchronization Signal (SSS, green), ∙ Cell Specific Reference Signal (Pilots, black), ∙ Physical Control Format Indicator Channel (PCFICH, cyan), ∙ Physical Hybrid ARQ Indicator Channel (PHICH, magenta), ∙ Physical Downlink Control Channel (PDCCH, yellow), ∙ Physical Broadcast Channel (PBCH, red), ∙ Physical Downlink Shared Channel (PDSCH, gray). 3GPP-LTE supports 1.4, 3, 5, 10, 15 and 20 MHz channel

Time domain correlation with Tx 1 scattered pilots

−3

SSS PSS

Correlation power →

1.5

x 10

No CFO compensation 1

0.5

0

8000

8200

8400

8600

8800 9000 Samples →

9200

9400

9600

9800

Time domain correlation with Tx 1 scattered pilots Correlation power →

0.1 With CFO compensation 0.08 0.06 0.04 0.02 0

bandwidths, which correspond to 72, 180, 300, 600, 900 and 1200 actively used OFDM subcarriers [3].

8400

8600

8800 9000 Samples →

9200

9400

9600

9800

Time domain correlation with Tx 1 PSS 0.06

0.04

0.02

0

B. Algorithmic Issues

4000 −3

1.5 Correlation power →

A typical 3GPP-LTE receiver, whose main purpose is for communication, would use the PSSs and SSSs for initial timing synchronization and carrier frequency offset estimation. For timing synchronization, one can exploit the special signal structure of SSSs by means of reverse differential correlation, the PSSs by cross correlation or simply exploit the guard interval [4]. Estimation and tracking of the channel and timings can be done by using the scattered pilot structure. For positioning purposes, the timing synchronization used for communication is not sufficient, e.g., the plateau-like correlation results from guard interval detection. Thus, we developed new concepts for initial access and synchronization [5]. Another important issue is the inter cell inteference due to the intended frequency reuse in 3GPP-LTE of 1. The performance at the cell edge and also at the core of a cell is limited due to non-ideal correlation properties of the used PSSs codes. The authors in [6] showed that an absolute received power level difference greater than 17.8 dB between two base stations leads to a complete detection loss of the weaker received signal. Iterative receiver concepts are therefore used to mitigate inter cell interference effects and increase the detection probability [6]. A quite important issue, also for a positioning receiver, is carrier frequency offset (CFO) estimation. The scattered pilots are very sensitive to CFO due to the destroyed orthogonality of subcarriers. Fig. 4 shows the time domain correlation with the scattered pilots of Tx 1, based on a real measurement snapshot. The upper plot shows the result before any CFO compensation, where as the lower plot shows the result after

8200

Fig. 4. Correlation with scattered pilots before and after CFO compensation.


Fig. 3. 3GPP-LTE frame structure for the core subcarriers of the spectrum. Mapping of physical channels to resource elements: Primary Synchronization Signal (PSS), Secondary Synchronization Signal (SSS), Cell Specific Reference Signal (Pilots), Physical Control Format Indicator Channel (PCFICH), Physical Hybrid ARQ Indicator Channel (PHICH), Physical Downlink Control Channel (PDCCH), Physical Broadcast Channel (PBCH), Physical Downlink Shared Channel (PDSCH).

8000

x 10

5000

6000

7000 Samples →

8000

9000

10000

9000

10000

Time domain correlation with Tx 1 scattered pilots

1

0.5

0

Fig. 5. pilots.

4000

5000

6000

7000 Samples →

8000

Initial time domain correlation with Tx 1 PSS and Tx 1 scattered

compensating an estimated CFO of 56.302 kHz. It is clearly visible that the pilots are highly sensitive to CFO, especially if one takes a look at the maximum correlation power. Before CFO compensation it is around 0.0015 and after correction it is approximately 0.09, which is a factor of 60 higher. The PSSs, comprised of Zadoff-Chu codes, are more robust to high CFO and can be detected by means of cross correlation, even though the offset is several subcarrier spacings high. Fig. 5 shows the initial time domain correlation with the PSS and scattered pilots of Tx 1. Also note, that the peak power in the upper plot in Fig. 5 is much higher than in the lower plot and there are no peak ambiguities. We also discovered that the PSSs time correlation peaks are shifted by a factor of several dozens of samples in the presence of high CFO, see Fig. 6. Note, the time samples difference between the green

Time domain correlation with Tx 1 PSS 0.08 After CFO compensation New detected peak Peak before CFO compensation

0.07


0.06

0.05

0.04

0.03

0.02

0.01

0 5000

Fig. 6.

5100

5200

5300

5400

5500 5600 Samples →

5700

5800

5900

6000

Correlation with Tx 1 PSS before and after CFO compensation.

0.09 Tx 1 PSS correlation Tx 1 Pilot correlation

0.08


0.07 0.06 0.05 0.04 0.03

narrower than the one from the PSS due to higher occupied bandwidth. Therefore, we use the pilot structure for fine timing synchronization. The algorithm flow of our new initial acquisition is as follows: 1) Time domain correlation with known, and base station specific PSSs. Detect the strongest Tx, which is then further processed. 2) Go to the first OFDM symbol where pilots are located, estimate the integer CFO based on scattered pilots and compensate the CFO. The frame structure is a-priori known, thus we can easily detect this first symbol. 3) Resynchronize by time correlation with the one specific PSSs of the detected strongest Tx. 4) Estimate the fractional CFO based on the pilots and compensate for it. 5) Estimate the fractional timing and channel frequency response based on the pilots. The resulting channel frequency response can be further processed with multipath-component estimation algorithms like SAGE and ESPRIT. 6) Collect all estimated parameters, like phase information, power, channel information and CFO and reconstruct the frame of the strongest Tx. 7) Subtract the generated reference frame from the incoming baseband data stream. Loop over to point 1 of the algorithm and try to detect the next strongest Tx. The acquisition is finished when all assumed Tx are processed. C. Experimental System Setup and Parameters

0.02 0.01 0

75

Fig. 7.

80

85 Samples →

90

95

PSS and scattered pilot correlation lobe for Tx 1.

peak and the red one, where as the green one represents the initial detection without CFO compensation. This fact does not hurt a communications receiver as long as this shift is within the guard interval and interference from the previous OFDM symbol is small, but it has a severe impact on a positioning receiver. Thus, after the initial integer CFO estimation, we do a timing resynchronization to get the true correlation peaks, see the red peak in Fig. 6. We also verified this during integration measurements of DLR’s LTE positioning test bed. It is also worth mentioning that we use the narrow-band PSSs and SSSs for initial detection only and not for fine timing synchronization. Fig. 7 shows the time domain correlation lobe width of the narrow band PSS and the scattered pilot of Tx 1, based on a real measurement snapshot. For visualization purposes we shifted the lobes on the time axis to clearly show the difference. The correlation lobe of the pilots is much

In this section we briefly summarize the system parameters and the general setup for the positioning measurements. Tab. II shows the main system parameters for DLR’s positioning test bed [7]. The test bed consists of four generic transmitters, realized in two FPGA boards, see Fig. 8. The predefined 3GPPLTE synchronization frames for each transmitter are emitted periodically. The test bed operates at a center frequency of 2.45 GHz. The radio frequency (RF) front end (FE) is comprised of commercial-off-the-shelf mixers, filters and amplifiers. The transmit antennas are near omni-directional. At the receiver we convert the RF signal directly down to baseband using a WiFi FE from Maxim Integrated Circuits. The analog inphase and quadrature components are then sampled with a data grabber and stored on an internal disk. The data grabbing unit consists of a common workstation with a build in high performance FPGA board with a PCI Express interface and a fast solid state drive (SSD). The stored raw samples are then further processed in MATLAB where we apply our new acquisition and channel estimation algorithms. Thus, a seamless integration of algorithms coming directly from simulation into a real hardware test bed is possible. Note, that the subcarrier spacing here is approximately 19 kHz compared to 15 kHz as defined in 3GPP-LTE [8]. Thus, the PSSs and the SSSs occupy around 1.4 MHz instead of 1 MHz. We use the scattered pilot structure as defined

60 MHz IF signal

2.4 GHz RF signal

FPGA Baseband TX 1 Digital Upconverter

Mobile Radio Channel 1

WLAN Frontend 2.4 GHz ISM-Band

PA Mobile Radio Channel 2

Sync

Analog I/Q

Control PA Mobile Radio Channel 3

Synchronization (Sync)


FPGA with PCI Express: Sampling Data transfer

PA Mobile Radio Channel 4

Sync FPGA Baseband TX 4 Digital Upconverter

Storage: SSD

Raw TDoA estimation

MATLAB: Acquisition Channel estimation TDoA estimation

PA

Additional channel information

Overview of DLR’s 3GPP-LTE positioning test bed with transmitters and receiver.

TABLE II S YSTEM PARAMETERS FOR MEASUREMENT

TX 1

TX 2 27.8 m LOS

Parameter type

Value

OFDM core symbol length Cyclic prefix length Sampling frequency Subcarrier spacing 𝑓𝑠𝑐 Pilot subcarrier spacing, same Tx Minimum pilot subcarrier spacing, different Tx Power Tx 1 Power Tx 2 Power Tx 3 Power Tx 4

1024 samples 144 samples 20 MHz 19.53 kHz 16 subcarriers 4 subcarriers 28.8 dbm 29.3 dbm 28.5 dbm 28.9 dbm

TX 3

Ground Truth Point Nr. 53

x x x x x x x x ….. x x x x… x x x x ………………x x x

Ground Truth Point Nr. 1

Door

17.2 m LOS

Fig. 8.

Workstation

ADC I/Q


TX 4

in 3GPP-LTE, where each transmitter Tx uses a different initialization for the random sequence generator. The pilots of different transmitters are therefore orthogonal in frequency, due to different subcarrier mapping, and in time, due to code division. The spatial length of one sample results in 𝜆 = 𝑐𝑓0 ≈ 3 ⋅ 108 ms ⋅ 20 1MHz = 15 m. The transmit power is set to around 750 mW to penetrate the highly attenuating windows at the DLR site. Fig. 9 shows an overview of the measurement site at the DLR’s Institue of Communications and Navigation. We placed four transmitters outside, where Tx 1 and Tx 2, as well as Tx 3 and Tx 4 are pairwise synchronized. The distances between the transmitters are 27.8 m and 17.2 m respectively, to show the dimension of the setup. We used one specific corridor where we measured 53 ground truth points, see the crossings in Fig. 9. The accuracy of these points is better than 5 cm and the LOS distance between point 1 and point 53 is 19.6 m. Note, that the rooms, numbered with 001 to 006 are typical office rooms, with whiteboards mounted on the wall which reflect and highly attenuate the incoming signals. The windows facing the transmitters already account for a approximately 30 dB loss, due to the metal coating. Fig. 10 shows an image of the corridor facing Tx 3, with the used omnidirectional receiving

Fig. 9.

Sketch of the measurement site.

antenna and the ground truth markings on the floor. On the left side of this corridor are the offices number 001 to 006. The upper part of Fig. 11 shows the true TDoA values for each ground truth point. The transmitters are pairwise synchronized, thus we get only two TDoA values instead of three. The lower part of Fig. 11 shows the histograms of both TDoA values for this measurement site. For Tx 1 and Tx 2 we get a uniform distribution because we move along the direct axis between those two transmitters. The same does not apply for Tx 3 and Tx 4 due to almost perpendicular movement to the direct axis. Also note that TDoA values are always smaller than 15 m, which means that all estimates are below one sample. The measurement scenario therefore actually relates to a femto-cell setup. Additionally to the TDoA measurement, we recorded GPS pseudo ranges for the same ground truth points. The measurements were taken with an u-blox EVK-6T GPS receiver

Number of visible GPS satellites

6

5

4

3

2

1

RX Antenna

0 0

Fig. 12.

10

20

30 Track point

40

50

Number of visible GPS satellites for each track point.

[9]. To obtain better signal conditions, some of the office windows where opened. However, the carrier to noise ratio of the received signal, is usually less than 20 dB-Hz.

Ground Truth Points

IV. E XPERIMENTAL R ESULTS

Fig. 10.

Image of corridor facing Tx 3.

15 Tx 1 - Tx 2 Tx 3 - Tx 4

TDoA [m]

10 5 0 -5 -10 -15

5

10

15

20 25 30 35 Ground truth point number 

0.2

40

45

0.25 Tx 3 - Tx 4 Normalized count

Normalized count

Tx 1 - Tx 2 0.15 0.1 0.05 0 -20

50

-10

0 TDoA [m]

10

20

0.2 0.15 0.1 0.05 0 -10

-8

-6 TDoA [m]

-4

-2

Fig. 11. True TDoA values for each ground truth point and corresponding histograms.

Several approaches and algorithms exist to solve the navigation equation for stand-alone GNSS or cellular, and hybrid solutions. This section considers an extended Kalman filter (EKF) for multisensor fusion of the GPS and LTE signals. Especially, this section shows the benefit of using the LTE network to obtain better positioning results. Multisensor fusion with the EKF uses directly the GPS pseudo ranges and TDoAs of the LTE system as observed signals. The track and the measurements setup were already described in Sec. III-C where we consider a straight walk through a corridor in an office building. The tracking considers a random walk with an initial speed of 1/3 m/s where the starting position is known. For calculating a position fix with GPS, more than 3 GPS signals are necessary. Therefore, Fig. 12 shows the number of available satellites for each track point. In the most of the cases less than 4 satellite signals are available. Additionally, these available signals are usually corrupted by non-line-ofsight and multipath errors. Thus, an accurate position estimate with GPS in indoor-scenarios is usually impossible. Fig. 13 shows the office building with the true track indicated by the green markers and the position estimations by the EKF. The magenta markers indicate the position estimate with the available GPS pseudo ranges. The blue markers show the tracking with LTE only and the red markers the fusion of GPS and LTE. By tracking with GPS only, an accurate positioning estimation in this indoor-scenario is impossible. Hence, Fig. 14 shows the positioning error in meters versus the track points. If we track with GPS only, the positioning error increases linearly, because the tracking of the position stays almost on the same position as shown in Fig. 13. Fig. 14 shows also the positioning error of the u-blox receiver, which increases

BS1

the GPS positioning by using an multipath error model as described e.g. [10]. Additionally some error models for the TDoA error will be considered e.g. [11].

BS2

y [m]

30

V. S UMMARY

BS3

20

10

0

True track EKF LTE only EKF GPS only EKF LTE and GPS

0

10

20 x [m]

BS4

30

ACKNOWLEDGMENT

40

Fig. 13. Floor plan of the office building in meter and overview of the tracked positions with the extended Kalman filter with GPS, LTE and LTE+GPS. 70

50 Positioning error [m]

This work has been performed in the framework of the FP7 project ICT-248894 WHERE2 [12] and the FP7 project GRAMMAR, grant agreement no. 227890) [13] which are partly funded by the European Union. Research leading to these results was also funded from the DLR project Galileo Advanced Applications (GalileoADAP). R EFERENCES

60

EKF LTE only EKF GPS only EKF LTE and GPS u−blox receiver

40

30

20

10

0 0

In this paper we have evaluated mobile radio based positioning using system parameters similar to those of 3GPP-LTE. We have theoretically evaluated the achievable performance for an outdoor-to-indoor environment, indicating that for the considered scenario the positioning accuracy is below 1-2 m. However, an experiment showed a positioning accuracy which is in the order of several meters. The discrepancy is most likely caused by multipath propagation, which has not been considered in our theoretical evaluations.

5

Fig. 14.

10

15

20

25 30 Track point

35

40

45

50

Positioning error in meter for each track point.

up to 70 m. The positioning estimation errors of the u–blox receiver and the GPS EKF are caused by non-line-of-sight and multipath effects. Additionally, the u–blox receiver cannot be initialized by the correct starting position. This causes already an positioning error of 23 m at the starting position. By tracking the position with LTE only, we obtain an positioning error of at most 7 m. Using the fusion of GPS and LTE, the positioning error can be slightly decreased. The fusion engine of the EKF weights in this scenario the GPS signals lower than the LTE signals. Thus, the EKF mainly follows LTE track but is corrected in vertical direction and reduces the maximum positioning error down to 5 m. However, the HDF is optimized for this scenario. In further investigations a more general solution has to be derived. Furthermore, we will concentrate on reducing the error in

[1] S. B. Weinstein and P. M. Ebert, “Data transmission by frequency division multiplexing using the Discrete Fourier Transform,” IEEE Transactions on Communications Technology, vol. 19, no. 5, pp. 628– 634, Oct. 1971. [2] S. M. Kay, Fundamentals of Statistical Processing — Estimation Theory. Prentice Hall, 1993, ISBN 0-13-345711-7. [3] LTE; Evolved Universal Terrestrial Radio Access (E-UTRA); Base Station (BS) radio transmission and reception (3GPP TS 36.104 version 8.6.0 Release 8), ETSI, July 2009, ETSI TS 136 104 V8.6.0. [4] J.-J. van de Beek, M. Sandell, and P. ola Brjesson, “ML estimation of time and frequency offset in OFDM systems,” in IEEE Transactions on Signal Processing, vol. 45, Juli 1997, pp. 1800–1805. [5] C. Mensing, S. Plass, and A. Dammann, “Synchronization algorithms for positioning with OFDM communications signals,” in 4th Workshop on Positioning, Navigation and Communication (WPNC), March 2007, pp. 205 – 210. [6] E. Staudinger and C. Gentner, “TDoA subsample delay estimator with multiple access interference mitigation and carrier frequency offset compensation for OFDM based systems,” in 8th Workshop on Positioning Navigation and Communication (WPNC), April 2011, pp. 33 – 38. [7] E. Staudinger, C. Klein, and S. Sand, “A generic OFDM based TDoA positioning testbed with interference mitigation for subsample delay estimation,” in Proceedings of 8th International Workshop on MultiCarrier Systems & Solutions (MC-SS 2011), Herrsching, Germany, May 2011. [8] J. Zyren and W. McCoy, “Overview of the 3GPP long term evolution physical layer,” Freescale, Tech. Rep., 2007. [9] u-blox 6 precision timing evaluation kit EVK-6T, http://www.ublox.com/en/evaluation-tools-a-software/gps-evaluation-kits/evk6t.html. [10] M. Khider, T. Jost, , E. Abdo, P. Robertson, and M. Angermann, “Bayesian multi-sensor navigation incorporating pseudo-ranges and multipath model,” in IEEE/ION PLANS, Indian Wells/Palm Springs, California, USA, May 2010. [11] W. Wang, T. Jost, C. Mensing, and A. Dammann, “ToA and TDoA error models for NLoS propagation based on outdoor to indoor channel measurement,” in Proceedings of the IEEE Wireless Communications and Networking Conference (WCNC) 2009, Budapest, Hungary, Apr. 2009. [12] EU FP7 Project WHERE2 (Wireless Hybrid Mobile Radio Estimators – Phase 2), http://www.ict-where2.eu. [13] EU FP7 Project GRAMMAR (Galileo Ready Advanced Mass MArket Receiver), http://www.gsa-grammar.eu.