Vector-Tracking-Loop of GNSS Receivers





01.08.2009 - 31.07.2011


Konkurk University, Seoul, Southkorea

 Projectleader at the Institute

Jong-Hoon Won



Recent success of digital processing technology provides a powerful tool to implement a software-based GNSS receiver that ideally issued many interesting application ideas. Instead of DLL, PLL and FLL that are widely used in a traditional GNSS signal tracking, the easy configurable property of software-based GNSS receiver has motivated many new improved signal tracking methods to process GNSS signals to improve the tracking performance in efficient ways.

A vector-based GNSS signal tracking (simply vector-tracking loop; VTL) is known as one of the most advanced techniques in this field. In the vector-tracking mode individual tracking loops are eliminated and effectively replaced by a global feedback path from a navigation filter to each local channel to make a loop closure, thereby enabling interactions between channels, and increasing the signal tracking sensitivity as well as anti-jamming robustness, especially when integrated with external sensors (e.g., inertial sensors).

A Vector-Tracking Loop (VTL) is an expansion of a Vector-Delay-Lock-Loop (VDLL) to the full use of code and carrier while the VDLL in strict sense uses only code delay. At the same time a VTL has a similar architecture with a deep coupling system except for the use of external sensor aiding information. The vector-tracking mode is based on the fact that the pseudorange measurement is not directly observable. Instead, the received signal is a single scalar observable, and each satellite-received signal component has two layered nonlinearities: the satellite-received signal is a function of signal parameters (code delay, Doppler, carrier phase and so on) that are also a function of the user’s navigation parameters, so that the received signal is a nonlinear function of the user’s navigation parameters. Therefore, there is a direct relationship between the user navigation parameters and the received signal measurement. As a result the user navigation parameters can be calculated directly from the received signals (i.e., one-step centralized filter approach). This is a quasi-optimal extension of the Extended Kalman Filter (EKF), however, it has some drawbacks in real-time implementation due to the fact that the signal processing channel should operate at a relatively higher rate than the navigation filter. Therefore the navigation filter cannot catch up with such a high operational speed. Moreover, the asynchronous property of each local signal tracking channel can be problematic.

In order to solve these problems, a two-step decentralized filtering approach is widely used. The basic idea of decentralized filtering approach is to determine efficient parallel processing structures that provide optimal estimates of the global system states. Designing a decentralized filter architecture generally seeks to replace the centralized architecture with a two-stage distributed architecture, i.e. a bank of local processors (slave filters) that process the measurements from the local sensors, followed by a global fusion processor (a master filter) that provides a global optimal solution by using the local processors outputs. In this case the error covariance matrices from local filters can be used when estimating the global system’s optimal state vector.

It is noted that the main advantages of vector-tracking loops (i.e., robustness, sensitivity, anti-jamming, etc.) are obtained from the fact that the system has a global feedback path to each local channel to have efficient interactions between channels not from the construction of nonlinear measurement equations of received signals in terms of navigation solutions. 

In fact, the advantages of VTL are obtained from mainly two facts:

* the efficient use of the global feedback path from the navigation filter to Numerically-Controlled-Oscillator (NCO) of each channel through the Line-of-Sight (LOS) projector

* the efficient use of the advanced form of signal tracking Kalman Filter (KF) instead of the conventional loop filters.


Several different forms of filtering methods have been proposed for the local channel filter design mainly in the field of GNSS and Inertial Navigation System (INS) deeply coupled integration systems. Actually, these design methodologies focused on a single channel tracking Kalman Filter (KF), where a local Kalman pre-filter was allocated to each signal tracking channel to process signal measurements (baseband I and Q or discriminator outputs) and then to provide residual estimates of signal parameters (i.e., code delay, carrier phase and Doppler frequency), that are converted to range and range-rate, to the (global) navigation filter (or fusion algorithm). Also an appropriate error covariance of the local filter output should be provided to the navigation filter for the system optimality.

Depending on how the measurement equation is constructed, they can be grouped into the following two distinct approaches:

* the direct use baseband Inphase (I) and Quadrature (Q) values

* the use of nonlinear discriminator outputs  

Beside of these, several different combinations of state variables can be used (e.g., signal parameter errors or pseudorange and range-rate errors including ionospheric delay errors, etc.). It was reported that a 7 dB gain in Carrier-to-Noise ratio (C/N0) could be obtained from the vector tracking loop. An interesting question from this point is, how big gains can be achieved from the loop closure by the global feedback path, and from the efficient use of the KF in the signal tracking loop.


In recent years GNSS receivers have been widely used in an autonomous vehicle navigation system as a main sensor with external independent sensors (wheel-speed, heading sensor, digital map, inertial sensors, etc.). A GNSS receiver, especially in urban canyons, has a problem of availability due to signal blockage and multipath, thereby causing the C/N0 drops in a standard receiver, which leads to the fact that the loss-of-lock occurs in a single or even in all channels of the receiver. Even if no complete loss-of-lock happens, the accuracy and integrity of the satellite navigation system is degraded.

One of the standard solutions for this problem is to perform sensor integration at system level, which provides higher robustness and integrity to the receiver. The usual way to implement this sensor integration is based on a “loose or tight coupling approaches” that are integrating the GNSS outputs (x, y, z, ∆t or raw observables) with the output from other sensors in a navigation filter, which is usually running on a microprocessor outside the GNSS receiver. The problem of these approaches is that the tracking sensitivity of a GNSS receiver is not enhanced because of the absence of feed-back loops from the navigation Kalman filter to the receiver digital processing section.

A more advanced concept to integrate the GNSS receiver with external sensors is called “deep coupling” technique. This technique was studied and developed by the US army to enhance jamming resistance of military GNSS user equipments. In available (non-classified) literature it was estimated that an anti-jamming margin of 15 - 20 dB could be reached with deep coupling technique. This means the GNSS signal can be tracked (and may be acquired) down to around 15 dB-Hz of C/N0. The deep coupling concept is currently not used in civil navigation systems (at least not commonly). From the stand-point of sensitivity (without a loss of lock of signal in degraded signal environment), re-acquisition in a short time, and autonomous vehicle sensor-based navigation (during the time intervals with a complete loss-of-lock), the deep coupling concept is without any doubt the most promising approach. It is clearly the most challenging sensor fusion concept. 

This deep coupling is tightly related to the vector-tracking-loop because the vector-tracking-loop has a very similar architecture as the deep coupling method except for the use of external aiding data (e.g., INS) of the deep coupling system. Therefore, this Technical Note starts from various coupling methods for the easy understanding of difference of various methods. 

As well known in navigation community GNSS and INS exhibit complementary features and can be efficiently integrated to yield a more robust performance. Over the years, the GNSS/INS integration has advanced from system level to deep inside of software/hardware level.

The first level of integration method treats both GNSS and INS as independent navigation systems, combining their position and velocity information, so-called Loose Coupling (LC) integration, where a Kalman filter is used to integrate the navigation states. This requires only the design of integration Kalman filter in position and velocity domain, thus causing the performance degradation because the exact covariance of the navigation parameters from the two systems is not known. And also position computations in GNSS receiver require a minimum of four visible satellites, which limits the use of this integrated system. The loose coupling is also called as cascaded INS/GNSS approach or position level blending algorithm in some contexts. In this integration, the INS is a main navigation system where the GPS is used only to aid the INS by resetting the navigation states.

The second level of integration method, so-called Tight Coupling (TC) integration, uses the pseudoranges and range-rates from a GNSS receiver instead of position and velocity for integrating with the INS output. This method can thus eliminate the unknown covariance problem in the loose coupling integration and also can be even when less than four (minimum) satellite visibility. This is because the tight coupling integration method regards each independent channel in a GNSS receiver as one complete sensor. The tight coupling is also called as close coupling or (pseudo) range level blending algorithm in some contexts. In this integration, the GNSS is a main navigation system where the INS is used only to aid the GNSS by interpolating two consecutive navigation states of GNSS.

The third level of integration method, so-called Ultra-Tight Coupling (UTC) integration, uses baseband I and Q values. Unlikely to loose and tight coupling integrations where the tracking loops of the GNSS receiver are controlled by the individual channels, in most of ultra-tight coupling integration the navigation filter provides the feedback value to NCOs of each channel. The main advantage of the ultra-tight coupling integration is that the carrier tracking loop bandwidth can be significantly reduced due to the fact that the INS Doppler-aiding successfully removes most of signal dynamics in the GNSS receiver’s signal tracking loop. This improves the quality of the measurements as well as the anti-jamming properties of the GNSS signals. The efficiency of this integration depends on the quality of the Doppler estimates derived from the INS. In the ultra-tight coupling integration system, the design of a signal processing block (specifically, a signal tracking loop), which has a feedback path from the navigation (and/or integration) filter, is needed to be emphasized. The ultra-tight coupling is also called as tight coupling in some contexts.

Besides of these, Deep Coupling integration method represents a GNSS receiver tracking loop using a vector-based processing method aided by inertial measurements. In wide sense, deep coupling may refer to the use of vector-based processing for GNSS signal tracking loops excluding externally aided inertial measurements. For another example, Vector-Delay-Lock-Loop (VDLL) represents a large internal GNSS signal tracking and navigation filter to calculate the user’s navigation information (position and time) directly from the incoming signals. Finally, inertial aiding represents any aiding to GNSS receiver, not limited to tracking loop, but also including acquisition and integer search aiding.


This figure shows top-level configurations of the three different GNSS/INS integration architectures focusing on the GNSS receiver’s internal signal processing functionalities. In the figure, the blue, green and red lines represent loose, tight, and ultra-tight coupling integration methods, respectively. Note that only ultra-tight coupling integration has a feedback path from GNSS/INS blending algorithm block (or navigation filter) to the GNSS receiver signal tracking loops. When other sensors are available, we can integrate them into the blending algorithm. As the level of integration moves more towards the deep inside of the system, the correlations in the raw measurements become less, i.e., coming to agreement with an uncorrelated additive Gaussian white noise property, therefore the sub-optimal property of integrated systems is improved.