This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (
Deadreckoning (DR) algorithms, which use selfcontained inertial sensors combined with gait analysis, have proven to be effective for pedestrian navigation purposes. In such DR systems, the primary error is often due to accumulated heading drifts. By tightly integrating global navigation satellite system (GNSS) Doppler measurements with DR, such accumulated heading errors can usually be accurately compensated. Under weak signal conditions, high sensitivity GNSS (HSGNSS) receivers with block processing techniques are often used, however, the Doppler quality of such receivers is relatively poor due to multipath, fading and signal attenuation. This often limits the benefits of integrating HSGNSS Doppler with DR. This paper investigates the benefits of using Doppler measurements from a novel direct vector HSGNSS receiver with pedestrian deadreckoning (PDR) for indoor navigation. An indoor signal and multipath model is introduced which explains how conventional HSGNSS Doppler measurements are affected by indoor multipath. Velocity and Doppler estimated by using direct vector receivers are introduced and discussed. Real experimental data is processed and analyzed to assess the veracity of proposed method. It is shown when integrating HSGNSS Doppler with PDR algorithm, the proposed direct vector method are more helpful than conventional block processing method for the indoor environments considered herein.
A wide range of commercial applications such as emergency services and cell phone locationbased services (LBS) have driven the development of pedestrian navigation technology over the past several years. With a demand for low cost and high reliability, attention has been given to using additional sensors or devices such as WiFi, inertial sensors and ZigBee radios to integrate with GNSS receivers. In particular, inertial sensors that can provide DR information have proven to be of great potential [
On the other hand, GNSS receivers are often used together with PDR algorithms due to the fact that their errors are not accumulated. Many researchers have investigated integrating PDR algorithms with global positioning system (GPS) receivers. The feasibility and performance of using a lowcost motion sensor integrated with GPS and differential GPS (DGPS) was assessed in [
In order to enhance the performance of the GNSS receiver, a typical method is to increase the coherent integration, as shown in [
With this in mind, the major objective of this paper is to investigate a new method of generating HSGNSS Doppler measurements with the goal of improving PDR implementation in certain degraded signal scenarios. A direct vector processing method is thus proposed and developed. First, the velocity maximum likelihood estimate (MLE) is obtained. Then, Doppler measurements are generated based on such velocity MLE. The advantage of this approach is its reliability in harsh indoor environments where line of sight (LOS) and/or nonLOS (NLOS) signals are present. Subsequently, the benefit of these measurements for improving PDR algorithms indoors is investigated. The methodology proposed here is analyzed based on the indoor signal and multipath models, which are intrinsically related with the distribution of multipath statistics. Real experimental data is then presented to further verify the effectiveness of the proposed methodology.
The contributions of the paper are twofold. First, a new direct vector processing receiver architecture is introduced and developed, which is shown to provide a more reliable velocity solution as well as Doppler measurements. Second, by using the new Doppler measurements integrated with PDR, the results are shown to improve the horizontal velocity accuracies by factors of more than 9% over the tradition implementation. Thus the effectiveness and benefits of the proposed Doppler estimation method are demonstrated and validated.
The paper is organized as follows: in Section 2, the signal and multipath models are introduced. After reviewing the architecture of conventional HSGNSS receivers, the proposed direct vector receiver is introduced. Then the velocity and Doppler estimation with direct vector processing in indoors are discussed in detail. In Section 3, the HSGNSS/PDR tight integration algorithm used in this paper is introduced. In Section 4, real indoor data is processed and analyzed. PDRonly solution, HSGPS/PDR tight integration with conventional Doppler and proposed Doppler measurements solutions are shown, compared, and discussed. Finally, conclusions are drawn in Section 5.
In this section, an indoor signal and multipath model is first introduced. The model is used to analyze how indoor multipath signals affect conventional HSGNSS Doppler estimation. After that, the proposed direct vector receiver architecture is introduced and discussed with comparison to the conventional HSGNSS receiver.
The environment considered herein is indoors with dense multipath, where the multipath delay spread is usually smaller than one chip duration, or equivalently, the coherence bandwidth is much larger than the signal bandwidth (spreading code bandwidth in GNSS case). Under this scenario, a nonfrequency selective channel or flatfading channel is usually assumed which implies the multipath timedelay is nonresolvable [
Once the radio frequency signal is received by the antenna, the receiver downconverts it to near baseband. At this point, the general complex signal envelope can be expressed as:
In
From
Having presented the basic signal model with dense multipath, attention is now given to how this signal is handled within a GNSS receiver, and how it affects the conventional HSGNSS Doppler estimation. The conventional block processing technique for Doppler measurement is discussed in [
If the coherent integration interval uses
In
In the vector form, the Doppler MLEs for all available satellites will be:
From
When a channel is time varying, the so called “Doppler spread” is sometimes introduced [
In
The architecture of a conventional high sensitivity GNSS software receiver that uses block processing technique is illustrated in
Direct vector receiver architecture is thus proposed in
As for conventional high sensitivity GNSS receivers, the velocity estimation is only based on the Doppler MLEs, which discards some information before final solution is made. Another major benefit of the direct vector receiver is that weighting is automatically performed according to the received signal strength [
It is known that the measured Doppler frequency from GNSS receivers has the following relationship with the user velocity:
From
In order to project the signal power from the Doppler domain to the velocity domain, the relationship between the Doppler to the velocity can be used. For example, a small offset in the Doppler will cause a small offset in the velocity, these two terms are linearly related as shown in
In
In order to visualize the effect of multipath on velocity estimation, it is convenient to first consider the two dimensional case. Assuming vertical velocity and clock drift are already known or constrained, the Doppler offset is then only related with two horizontal velocity offsets (Δ
As more satellites are considered, the situation tends to that shown in
The velocity estimation procedure shown in
On the other hand, as the number of satellites tracked increases, the LOS signal power in the velocity domain is accumulated without loss, while the NLOS signal power in velocity domain is usually dispersed. The reason is that the projection matrix
The above section discusses the power projection in two dimensional spaces. In the following, the same principles are extended to the real scenario where four velocity states need to be estimated. Assuming there are N
Each hyperplane is associated with the LOS signal power of
For example, the peak powers associated with each hyperplane of NLOS signal are
The velocity MLE can be obtained by using the
In this section, the system model for the HSGNSS/PDR tight integration is first introduced. The measurement or observation models are then presented. Following this, the integration performance of using conventional and the proposed Doppler measurements with a PDR algorithm will be assessed in the next section.
The system state vector for the PDR filter is shown in
The GNSS receiver clock bias is not present, since in this paper only the Doppler measurements are used to update the integration filter. In turn, this is because the paper focuses on assessing the benefit of the proposed Doppler measurements for PDR.
The system dynamic model of the pedestrian's position follows the equations of a classical PDR mechanization and is given by:
The velocities and the heading are further modeled as random walk processes:
Similarly, the clock drift state is modeled as:
In Equations
It is noted that discrepancies between PDRderived velocity and GNSSderived velocity are expected. In particular, due to the repetitive nature of the human gait, Doppler measurements typically exhibit oscillations over the course of a full gait cycle. In contrast, by its very nature, PDR velocities are effectively averaged over the course of a step and do not contain these oscillations, thus introducing an oscillatory discrepancy between the PDR and GNSSderived values. Correspondingly, these oscillations should be modeled to properly integrate the velocity information based on Doppler measurements.
For coping with these oscillations over a step, the classical PDR has been modified in this work. Normally, the filter state is only propagated when a step is detected. In the proposed approach, however, Doppler measurements are used at each GNSS measurement epoch, typically at a higher rate than the step frequency. As such, the PDR filter is propagated at this higher rate as well. However, PDR observations are still only used when a step is detected. It is expected that this approach will average the oscillating effects sensed by Doppler measurements over one step. The advantage of this asynchronous measurement update is that the integrated system could track the rapid changes of the heading sensed by the IMU or other heading sensors.
On the other hand, there are also some disadvantages with this new integration method. First, the computational load is increased because the filter is propagated more frequently. In this paper, a coherent integration time of 500 ms is used, such that the measurement update time for Doppler is 2 Hz, so computational load is not a major concern. Second, as the coherent integration time increases, the Doppler measurement actually conveys information about the average velocity (and thus attitude) during the integration interval. However, this type of averaged Doppler measurement is still usable since it can help to alleviate the long term heading drift of the PDR system.
Having introduced the system model of the integration filter, the following discusses the measurement models. There are two types of measurement updates for the proposed tight integration; from the PDR and from the GNSS Doppler.
The PDR sensor update is composed of step length updated and heading update. The measured step length and walking directions are related to the user's position and velocity through the following equations:
The measured GNSS Doppler is related to the pedestrian's velocity via
In
It is noted that PDR only provides information in the horizontal plane rendering the vertical velocity unobservable except with Doppler observations. To further constrain the vertical component of the pedestrian's position, barometer records could be used, although this was not done here. Similarly, indoors, pedestrians are mainly walking on flat surfaces and only change their elevation when climbing/descending stairs or when taking an elevator.
This section deals with real experimental data processing and analysis in order to assess the benefits of integrating conventional and proposed HSGNSS Doppler measurements with PDR sensors. First, the data collection is described briefly. Then the analysis and results are described. For simplicity, only GPS satellites are used, but it is expected that results would likely improve if a multiple GNSS constellation were used.
The experimental data was collected on the campus of the University of Calgary. The primary pieces of equipment were an NI frontend, a Novatel SPAN receiver and a LCI IMU [
The raw IF data was collected with the NI frontend at a rate of 5 Msps (complex). The reference trajectory is shown in
The trajectory contains several different indoor environments, and two spots are chosen to emphasize how Doppler from direct vector HSGPS outperforms that from conventional HSGPS (scenario A and scenario B). The pedestrian carries a backpack containing an antenna, the NovAtel SPAN system and LCI IMU. The cable from the backpack is connected to the NI frontend.
One of the benefits of the proposed direct vector processing is its autonomous weighting by power (equivalent to C/N_{0}). Ideally, if it is possible to get accurate enough C/N_{0} estimates in the conventional HSGPS, the results can be very close to the proposed approach. However, there are some difficulties for C/N_{0} estimation in weak signal and multipath conditions. As shown in
The sky plot of all available GPS satellites is shown in
To begin the test, the pedestrian walks in a circular path outside the building in order to align the inertial system. Then, the user walks through the inside of the building with periodic returns outside in order to maintain an accurate reference solution (details below). Finally, circular motion is repeated again at the end of the test in open sky scenario in order to facilitate backward processing of the data.
The raw IF data was processed using the GNSRxss™ software receiver [
GSNRxss™ with a coherent integration of 500 ms is used. With this coherent integration time, it is expected that a desirable predetection SNR can be obtained. After projecting the correlator outputs onto the velocity domain, the velocity powers are computed using
In
Once correlator outputs from the receivers are available, the MLE velocities are computed first. The resulting velocity domain power distributions in the two indoor environments are depicted in
In scenario A, it can be observed that around epoch 15 s, the total received power is the greatest and the dominant power is located very near the reference solution; the LOS signal appears to dominate the signal. However in the succeeding epochs, NLOS signals seem to dominate the received signal. The error statistics of the proposed direct vector and conventional HSGPS velocity solutions for this scenario are summarized in
Since the multipath statistics are environmentdependent, it is useful to show that the proposed algorithm works in various indoor environments. In
Comparing results of the above two scenarios, two major phenomena are observed. First, the accumulated power fluctuates more from epoch to epoch while indoors. Second, the location of the peak power in the velocity domain largely depends on the multipath statistics or environments.
In direct vector processing, the velocity MLE is first obtained, and then the corresponding Doppler measurements are computed. It is then convenient to assess the performance of the HSGPS/PDR tight integration using various types of observation. In the following paragraphs, the performance of the PDR only navigation solution and HSGPS/PDR tight integration with conventional and proposed Doppler measurements is assessed.
Here tight integration uses only Doppler measurements to update the user velocity and heading. No pseudorange measurements are used. In this way, the accumulating errors caused by inaccurate Doppler measurements will become more apparent. The other reason to choose HSGPS/PDR integration with conventional and proposed Doppler measurements is to make the results comparable. If the velocity MLE is directly integrated with PDR, it is then a loose integration scheme. This might obscure the benefits of the measurements and integration schemes. By using the same integration scheme, the benefits of better Doppler measurements will become evident.
For the DR algorithms, the step event is first detected by using pattern recognition techniques with a MEMS accelerometer [
The trajectories of PDR only and HSGPS/PDR tight integration with conventional and proposed Doppler measurements are shown in
Among all three navigation solutions, it can be seen that the navigation solution with the proposed Doppler measurements integrated with PDR is the nearest to the reference trajectory, which validates the effectiveness of the proposed method. In order to further show the error characteristics, the position and velocity errors are plotted as a function of time in
To further analyze the results, the cumulative histograms of the horizontal and vertical position errors are given in
The corresponding RMS error statistics are summarized in
These results suggest that direct vector HSGPS Doppler actually reduces such errors on average, which, as discussed below, improves heading determination. For this data set, the benefit from direct vector HSGPS Doppler is more significant in the north direction. To further illustrate the improvement, the mean velocity errors are shown in
In
In this paper, the combined indoor signal and multipath model is first introduced. How multipath affects conventional HSGNSS Doppler measurements is then discussed. After that direct vector receiver architecture is proposed and compared to conventional high sensitivity GNSS receivers. The conventional method performs well in most good scenarios, and the proposed method is the maximum likelihood extension to the conventional method, which will asymptotically approach the performance bound. How velocity and Doppler measurements are obtained with direct vector processing performing in indoor multipath environments is also investigated. In order to evaluate the benefits of Doppler estimated with the proposed direct vector processing method over conventional block processing method, Doppler estimated with both approaches are tightly integrated in a detailed navigation filter, which follows a PDR strategy. Comparisons are made between PDR only solution and integrated solutions. From the experimental results, the following conclusions can be drawn:
In indoor environments, the direct vector processing makes the best use of LOS and NLOS signal power for each satellite. Additionally, the mutual information between satellites is further used before arriving at a solution.
Velocity estimation by using direct vector HSPGS outperforms the conventional HSGPS for two indoor scenarios considered herein. The HSGPS Doppler measurements obtained by using direct vector processing are thus more reliable and helpful than conventional HSGPS Doppler measurements for heading estimation.
In weak signal and multipath conditions, the benefit of using conventional HSGPS Doppler measurements is very limited due to the fact that proper weighting cannot be correctly set. However, the proposed direct vector processing method autonomously weighs the navigation solution according to power (equivalent to C/N_{0}), which is actually the maximum likelihood estimate of the navigation solution, and performs no worse or better than the conventional approach.
By integrating the Doppler measurements from the proposed direct vector processing with PDR, the solutions show improvements in east and north velocity estimation of 16% and 9% for the indoor scenarios considered herein as compared to the conventional approach.
Currently only the GPS system in L1 band is used. In an ongoing project, the GLONASS signals in L1 and L2 bands will also be included. It can be expected that with increased number of satellites, the proposed Doppler will have a tendency to be closer to the true value under certain circumstances. Various indoor environments will be assessed in order to show the strength and weakness of the methods proposed in this paper. Regarding the integration, other heading sensors, such as magnetometers, and lowcost MEMS gyroscopes will also be included, and the benefits of using HSGNSS Doppler measurements in various indoor environments will be further assessed and analyzed.
Financial support from Research In Motion (RIM), the Natural Science and Engineering Research Council of Canada, Alberta Advanced Education and Technology and the Western Economic Diversification Canada is acknowledged.
Conventional high sensitivity GNSS receiver architecture.
Direct vector GNSS receiver architecture.
Power projection from Doppler domain to velocity domain (2D case).
Velocity MLE with channel distortion (2D case).
Primary equipments used in data collection. SPAN™ LCI IMU, CPT, and SPAN™ receivers are shown on the left; NI frontend is shown on the right.
Data collection actual trajectory, indoor scenarios, and corresponding C/N0 plots (one epoch is 0.5 s)—scenario A picture taken facing east, scenario B picture taken facing south.
Sky plots of all available GPS satellites.
Power distributions in east, north and up velocity domains (
Power distributions in east, north and up velocity domains (
Trajectories of various navigation solutions.
Position errors of various navigation solutions.
Velocity errors of various navigation solutions.
Cumulative histograms of horizontal and vertical position errors.
Interpolated heading errors.
Velocity RMS errors–indoor scenario A.
East (m/s)  0.14  0.56  75% 
North (m/s)  0.32  1.29  75% 
Up (m/s)  0.52  4.96  89% 
Velocity RMS errors–indoor scenario B.
East (m/s)  0.19  0.25  24% 
North (m/s)  0.32  1.16  72% 
Up (m/s)  0.41  2.43  83% 
Position and velocity RMS errors.

 

PDRonly  33.65  93.18  2.23  0.38  0.47  0.16 
PDR+HSGPS (conv.)  75.71  35.41  7.55  0.49  0.45  0.20 
PDR+HSGPS (dir.)  25.52  37.45  5.53  0.41  0.41  0.20 
Mean velocity errors.
 

PDRonly  0.02  −0.06  0.00 
PDR+HSGPS (conventional)  0.02  0.01  0.00 
PDR+HSGPS (direct)  −0.01  −0.01  0.00 