Imported: 13 Feb '17 | Published: 18 Jan '11
USPTO - Utility Patents
A method and system for a long range Radio Frequency (RF)-based identification, tracking and locating of objects. The method and system use a narrow bandwidth ranging signal(s), including VHF of lower frequency bands, which minimizes propagation loss and loss of accuracy of the RF locating signals. The method and system includes narrow bandwidth ranging signal multi-path mitigations processor, which further improves the track-locate accuracy. The signal is sent from a Master Unit(s) to a Tag. The signal traveling time is recorded and the distance between the Master(s) and the Tag is calculated. The method and system allow achieving a longer distance of the RF narrow bandwidth ranging signal penetration and an increased accuracy by using VHF bands in conjunction with the narrow bandwidth ranging signal multi-path mitigations processor. The techniques of Digital Signal Processing and Software-Defined Radio are used. The actual waveforms transmitted and received by the radios are defined by the software. The roles of the Master Unit(s) and the Tag can be reversed.
This application is a continuation of U.S. patent application Ser. No. 11/610,595, filed on Dec. 14, 2006, entitled METHODS AND SYSTEM FOR REDUCED ATTENUATION IN TRACKING OBJECTS USING RF TECHNOLOGY, which claims priority to U.S. provisional Patent Application No. 60/597,649 filed on Dec. 15, 2005, entitled METHOD AND SYSTEM FOR REDUCED ATTENUATION IN TRACKING OBJECTS USING MULTI-BAND RF TECHNOLOGY, which are incorporated by reference herein in their entirety.
This application claims priority to U.S. provisional Patent Application No. 61/103,270, filed on Oct. 7, 2008, entitled METHODS AND SYSTEM FOR MULTI-PATH MITIGATION IN TRACKING OBJECTS USING REDUCED ATTENUATION RF TECHNOLOGY, which is incorporated by reference herein in its entirety.
1. Field of the Invention
The present invention relates to method and system for a Radio Frequency (RF)-based identification, tracking and locating of objects, including RTLS (Real Time Locating Service).
2. Background of the Related Art
RF-based identification and location-finding systems for determination of relative or geographic position of objects are generally used for tracking single objects or groups of objects, as well as for tracking individuals. Conventional location-finding systems have been used for position determination in an open outdoor environment. RF-based, Global Positioning System (GPS), and assisted GPSs are typically used. However, conventional location-finding systems suffer from certain inaccuracies when locating the objects in closed (i.e., indoor) environments, as well as outdoors.
The indoor and outdoor location inaccuracies are due mainly to the physics of RF propagation, in particular, due to losses/attenuation of the RF signals, signal scattering and reflections. The losses/attenuation and scattering issues can be solved (see co-pending application Ser. No. 11/670,595) by employing narrow-band ranging signal(s) and operating at low RF frequencies, for example at VHF range or lower.
Although, at VHF and lower frequencies the multi-path phenomena (e.g., RF energy reflections), is less severe than at UHF and higher frequencies, the impact of the multi-path phenomena on location-finding accuracy makes location determination less reliable and precise than required by the industry. Accordingly, there is a need for a method and a system for mitigating the effects of the RF energy reflections (i.e., multi-path phenomena) in RF-based identification and location-finding systems that are employing narrow-band ranging signal(s).
As a rule, conventional RF-based identification and location-finding systems mitigating multipath by employing wide bandwidth ranging signals, e.g. exploiting wide-band signal nature for multi-path mitigation (see S. Salous, “Indoor and Outdoor UHF Measurements with a 90 MHz Bandwidth”, IEEE Colloquium on Propagation Characteristics and Related System Techniques for Beyond Line-of-Sight Radio, 1997, pp. 8/1-8/6). In addition, spatial diversity and/or antenna diversity techniques are used in some cases.
However, the spatial diversity may not be an option in many tracking-location applications because it leads to an increase in required infrastructure. Similarly, the antenna diversity has a limited value, because at lower operating frequencies, for example VHF, the physical size of antenna subsystem becomes too large. The case in point is the U.S. Pat. No. 6,788,199, where a system and method for locating objects, people, pets and personal articles is described.
The proposed system employs antenna array to mitigate the multi-path. The system operates at UHF in the 902-926 MHz frequencies band. It is well known that the linear dimension of the antenna is proportional to the wave length of an operating frequency. Also, the area of an antenna array is proportional to the square and volume to the cube of the linear dimensions ratio because in an antenna array the antennas are usually separated by ¼ or ½ wave length. Thus, at VHF and lower frequencies the size of the antenna array will significantly impact device portability.
On the other hand, because of a very limited frequency spectrum, the narrow bandwidth ranging signal does not lend itself into multi-path mitigation techniques that are currently used by conventional RF-based identification and location-finding systems. The reason is that the ranging signal distortion (i.e., change in the signal) that is induced by the multi-path is too small for reliable detection/processing in presence of noise. Also, because of limited bandwidth the narrow bandwidth receiver cannot differentiate between ranging signal Direct-Line-Of-Sight (DLOS) path and delayed ranging signal paths when these are separated by small delays, since the narrow bandwidth receiver lacks the required time resolution, which is proportional to the receiver's bandwidth (e.g., the narrow bandwidth has an integrating effect on the incoming signals).
Accordingly, there is a need in the art for a multi-path mitigation method and system for object identification and location-finding, which uses narrow bandwidth ranging signal(s) and operates in VHF or lower frequencies as well as UHF band frequencies and beyond.
The present invention relates to a method and system for a Radio Frequency (RF)-based identification, tracking and locating of objects, including Real Time Locating Service (RTLS) that substantially obviates one or more of the disadvantages of the related art. The proposed method and system use a narrow bandwidth ranging locating signal(s). According to an exemplary embodiment, RF-based tracking and locating is implemented on VHF band, but could be also implemented on lower bands (HF, LF and VLF) as well as UHF band and higher frequencies. It employs multi-path mitigation method including techniques and algorithms. The proposed system can use software implemented digital signal processing and software defined radio technologies. Digital signal processing can be used as well.
The system of the exemplary embodiment can be constructed using standard FPGAs and standard signal processing hardware and software at a very small incremental cost to the device and overall system. At the same time the accuracy of the RF-based identification and location-finding systems that are employing narrow-band ranging signal/s can be significantly improved.
The transmitters and receivers for narrow bandwidth ranging/locating signal, for example VHF, are used to identify a location of a person or an object. Digital signal processing (DSP) and software defined radio (SDR) technologies can be used to generate, receive and process a narrow bandwidth ranging signal(s) as well as perform multi-path mitigation algorithms. The narrow bandwidth ranging signal is used to identify, locate and track a person or an object in a half-duplex, full duplex or simplex mode of operation. The Digital signal processing (DSP) and software defined radio (SDR) technologies are used in the multi-path mitigation processor to implement multi-path mitigation algorithms.
Additional features and advantages of the invention will be set forth in the description that follows, and in part will be apparent from the description, or may be learned by practice of the invention. The advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are intended to provide further explanation of the invention as claimed.
Reference will now be made in detail to the preferred embodiments of the present invention, examples of which are illustrated in the accompanying drawings.
The present invention relates to a method and system for RF-based identification, tracking and locating of objects, including RTLS. According to the exemplary embodiment, the method and system employs a narrow bandwidth ranging signal. The exemplary embodiment operates in VHF band, but can be also used in HF, LF and VLF bands as well as UHF band and higher frequencies. It employs multi-path mitigation processor. Employing multi-path mitigation processor increases the accuracy of tracking and locating implemented by a system.
The exemplary embodiment includes small, highly portable base units that allow users to track, locate and monitor multiple persons and objects. Each unit has its own ID. Each unit broadcasts an RF signal with its ID, and each unit is able to send back a return signal, which can include its ID as well as voice, data and additional information. Each unit processes the returned signals from the other units and, depending on the triangulation or trilateralization and/or other methods used, continuously determines their relative and/or actual locations. The preferred embodiment can also be easily integrated with products such as GPS devices, smart phones, two-way radios and PDAs. The resulting product will have all of the functions of the stand-alone devices while leveraging the existing display, sensors (such as altimeters, GPS, accelerometers and compasses) and processing capacity of its host. For example, a GPS device with the device technology describe herein will be able to provide the user's location on a map as well as to map the locations of the other members of the group.
The size of the preferred embodiment based on an FPGA implementation is between approximately 2×4×1 inches and 2×2×0.5 inches, or smaller, as integrated circuit technology improves. Depending on the frequency used, the antenna will be either integrated into the device or protrude through the device enclosure. An ASIC (Application Specific Integrated Circuit) based version of the device will be able to incorporate the functions of the FPGA and most of the other electronic components in the unit or Tag. The ASIC-based stand-alone version of the product will result in the device size of 1×0.5×0.5 inches or smaller. The antenna size will be determined by the frequency used and part of the antenna can be integrated into the enclosure. The ASIC based embodiment is designed to be integrated into products can consist of nothing more than a chipset. There should not be any substantial physical size difference between the Master or Tag units.
The devices can use standard system components (off-the-shelf components) operating at multiple frequency ranges (bands) for processing of multi-path mitigation algorithms. The software for digital signal processing and software-defined radio can be used. The signal processing software combined with minimal hardware, allows assembling the radios that have transmitted and received waveforms defined by the software.
Co-pending application Ser. No. 11/670,595 discloses a narrow-bandwidth ranging signal system, whereby the narrow-bandwidth ranging signal is designed to fit into a low-bandwidth channel, for example using voice channels that are only several kilohertz wide (though some of low-bandwidth channels may extend into a few tens of kilohertz). This is in contrast to conventional location-finding systems that use channels from hundreds of kilohertz to tens of megahertz wide.
The advantage of this narrow-bandwidth ranging signal system is as follows: 1) at lower operating frequencies/bands, conventional location-finding systems ranging signal bandwidth exceeds the carrier (operating) frequency value. Thus, such systems cannot be deployed at LF/VLF and other lower frequencies bands, including HF. Unlike conventional location-finding systems, the narrow-bandwidth ranging signal system described in co-pending application Ser. No. 11/670,595 can be successfully deployed on LF, VLF and other bands because its ranging signal bandwidth is far below the carrier frequency value; 2) at lower end of RF spectrum (some VLF, LF, HF and VHF bands), e.g., up to UHF band, conventional location-finding systems cannot be used because the FCC severely limits the allowable channel bandwidth (12-25 kHz), which makes it impossible to use conventional ranging signals. Unlike conventional location-finding systems, the narrow-bandwidth ranging signal system's ranging signal bandwidth is fully compliant with FCC regulations and other international spectrum regulatory bodies; and 3) it is well known (see MRI: the basics, by Ray H. Hashemi, William G. Bradley . . . —2003) that independently of operating frequency/band, a narrow-bandwidth signal has inherently higher SNR (Signal-to-Noise-Ratio) as compared to a wide-bandwidth signal. This increases the operating range of the narrow-bandwidth ranging signal location-finding system independently of the frequency/band it operates, including UHF band.
Thus, unlike conventional location-finding systems, the narrow-bandwidth ranging signal location-finding system can be deployed on lower end of the RF spectrum—for example VHF and lower frequencies bands, down to LF/VLF bands, where the multipath phenomena is less pronounced. At the same time, the narrow-bandwidth ranging location-finding system can be also deployed on UHF band and beyond, improving the ranging signal SNR and, as a result, increasing the location-finding system operating range.
To minimize multipath, e.g., RF energy reflections, it is desirable to operate on VLF/LF bands. However, at these frequencies the efficiency of a portable/mobile antenna is very small (about 0.1% or less because of small antenna length (size) relative to the RF wave length). In addition, at these low frequencies the noise level from natural and manmade sources is much higher than on higher frequencies/bands, for example VHF. Together, these two phenomena may limit the applicability of location-finding system, e.g. its operating range and/or mobility/portability. Therefore, for certain applications where operating range and/or mobility/portability are very important a higher RF frequencies/bands may be used, for example HF, VHF, UHF and UWB.
At VHF and UHF bands, the noise level from natural and manmade sources is significantly lower compared to VLF, LF and HF bands; and at VHF and HF frequencies the multi-path phenomena (e.g., RF energy reflections) is less severe than at UHF and higher frequencies. Also, at VHF, the antenna efficiency is significantly better, than on HF and lower frequencies, and at VHF the RF penetration capabilities are much better than at UHF. Thus, the VHF band provides a good compromise for mobile/portable applications. On the other hand in some special cases, for example GPS where VHF frequencies (or lower frequencies) cannot penetrate the ionosphere (or get deflected/refracted), the UHF can be a good choice. However, in any case (and all cases/applications) the narrow-bandwidth ranging signal system will have advantages over the conventional wide-bandwidth ranging signal location-finding systems.
The actual application(s) will determine the exact technical specifications (such as power, emissions, bandwidth and operating frequencies/band). Narrow bandwidth ranging allows the user to either receive licenses or receive exemption from licenses, or use unlicensed bands as set forth in the FCC because narrow band ranging allows for operation on many different bandwidths/frequencies, including the most stringent narrow bandwidths: 6.25 kHz, 11.25 kHz, 12.5 kHz, 25 kHz and 50 kHz set forth in the FCC and comply with the corresponding technical requirements for the appropriate sections. As a result, multiple FCC sections and exemptions within such sections will be applicable. The primary FCC Regulations that are applicable are: 47 CFR Part 90—Private Land Mobile Radio Services, 47 CFR Part 94 personal Radio Services, 47 CFR Part 15—Radio Frequency Devices. (By comparison, a wideband signal in this context is from several hundred KHz up to 10-20 MHz.)
Typically, for Part 90 and Part 94, VHF implementations allow the user to operate the device up to 100 mW under certain exemptions (Low Power Radio Service being an example). For certain applications the allowable transmitted power at VHF band is between 2 and 5 Watts. For 900 MHz (UHF band) it is 1 W. On 160 kHz-190 kHz frequencies (LF band) the allowable transmitted power is 1 Watt.
Narrow band ranging can comply with many if not all of the different spectrum allowances and allows for accurate ranging while still complying with the most stringent regulatory requirements. This holds true not just for the FCC, but for other international organizations that regulate the use of spectrum throughout the world, including Europe, Japan and Korea.
The following is a list of the common frequencies used, with typical power usage and the distance the tag can communicate with another reader in a real world environment (see Indoor Propagation and Wavelength Dan Dobkin, WJ Communications, V 1.4 7/10/02):
The proposed system works at VHF frequencies and employs a proprietary method for sending and processing the RF signals. More specifically, it uses DSP techniques and software-defined radio (SDR) to overcome the limitations of the narrow bandwidth requirements at VHF frequencies.
Operating at lower (VHF) frequencies reduces scatter and provides much better wall penetration. The net result is a roughly ten-fold increase in range over commonly used frequencies. Compare, for example, the measured range of a prototype to that of the RFID technologies listed above:
Battery consumption is a function of design, transmitted power and the duty cycle of the device, e.g., the time interval between two consecutive distance (location) measurements. In many applications the duty cycle is large 10× to 1000×. In applications with large duty cycle, for example 100×, an FPGA version that transmits 100 mW of power will have an up time of approximately three weeks. An ASIC based version is expected to increase the up time by 10×. Also, ASICs have inherently lower noise level. Thus, the ASIC-based version may also increase the operating range by about 40%.
Those skilled in the art will appreciate that the exemplary embodiment does not compromise the system long operating range while significantly increases the location-finding accuracy in RF challenging environments (such as, for example, buildings, urban corridors, etc.)
Typically, tracking and location systems employ Track-Locate-Navigate methods. These methods include Time-Of-Arrival (TOA), Differential-Time-Of-Arrival (DTOA) and combination of TOA and DTOA. Time-Of-Arrival (TOA) as the distance measurement technique is generally described in U.S. Pat. No. 5,525,967. A TOA/DTOA-based system measures the RF ranging signal Direct-Line-Of-Site (DLOS) time-of-flight, e.g., time-delay, which is then converted to a distance range.
In case of RF reflections (e.g., multi-path), multiple copies of the RF ranging signal with various delay times are superimposed onto the DLOS RF ranging signal. A track-locate system that uses a narrow bandwidth ranging signal cannot differentiate between the DLOS signal and reflected signals without multi-path mitigation. As a result, these reflected signals induce an error in the estimated ranging signal DLOS time-of-flight, which, in turn, impacts the range estimating accuracy.
The exemplary embodiment advantageously uses the multi-path mitigation processor to separate the DLOS signal and reflected signals. Thus, the exemplary embodiment significantly lowers the error in the estimated ranging signal DLOS time-of-flight. The proposed multi-path mitigation method can be used on all RF bands. It can also be used with wide bandwidth ranging signal location-finding systems. And it can support various modulation/demodulation techniques, including Spread Spectrum techniques, such as DSS (Direct Spread Spectrum) and FH (Frequency Hopping).
Additionally, noise reduction methods can be applied in order to further improve the method's accuracy. These noise reduction methods can include, but are not limited to, coherent summing, non-coherent summing, Matched filtering, temporal diversity techniques, etc. The remnants of the multi-path interference error can be further reduced by applying the post-processing techniques, such as, maximum likelihood (Vitterbi Algorithm), Kalman filtering (Kalman Algorithm), etc.
The exemplary embodiment can be used in systems with simplex, half-duplex and full duplex modes of operation. Full-duplex operation is very demanding in terms of complexity, cost and logistics on the RF transceiver, which limits the system operating range in portable/mobile device implementations. In half-duplex mode of operation the reader (often referred to as the “master”) and the tags (sometimes also referred to as “slaves” or “targets”) are controlled by a protocol that only allows the master or the slave to transmit at any given time.
The alternation of sending and receiving allows a single frequency to be used in distance measurement. Such an arrangement reduces the costs and complexity of the system in comparison with full duplex systems. The simplex mode of operation is conceptually simpler, but requires a more rigorous synchronization of events between master and target unit(s), including the start of the ranging signal sequence.
In present invention the narrow bandwidth ranging signal multi-path mitigation processor does not increase the ranging signal bandwidth. It uses different frequency components, advantageously, to allow propagation of a narrow bandwidth ranging signal. Further ranging signal processing can be carried out in the frequency domain by way of employing super resolution spectrum estimation algorithms (MUSIC, rootMUSIC, ESPRIT) and/or statistical algorithms like RELAX, or in time-domain by assembling a synthetic ranging signal with a relatively large bandwidth and applying a further processing to this signal. The different frequency component of narrow bandwidth ranging signal can be pseudo randomly selected, it can also be contiguous or spaced apart in frequency, and it can have uniform and/or non-uniform spacing in frequency.
The embodiment expands multipath mitigation technology. The signal model for the narrowband ranging is a complex exponential (as introduced elsewhere in this document) whose frequency is directly proportional to the delay defined by the range plus similar terms whose delay is defined by the time delay related to the multipath. The model is independent of the actual implementation of the signal structure, e.g., stepped frequency, Linear Frequency Modulation, etc.
The frequency separation between the direct path and multipath is nominally extremely small and normal frequency domain processing is not sufficient to estimate the direct path range. For example a stepped frequency ranging signal at a 100 KHz stepping rate over 5 MHz at a range of 30 meters (100.07 nanoseconds delay) results in a frequency of 0.062875 radians/sec. A multipath reflection with a path length of 35 meters would result in a frequency of 0.073355. The separation is 0.0104792. Frequency resolution of the 50 sample observable has a native frequency resolution of 0.12566 Hz. Consequently it is not possible to use conventional frequency estimation techniques for the separation of the direct path from the reflected path and accurately estimate the direct path range.
To overcome this limitation the invention uses a unique combination of implementations of subspace decomposition high resolution spectral estimation methodologies and multimodal cluster analysis. The subspace decomposition technology relies on breaking the estimated covariance matrix of the observed data into two orthogonal subspaces, the noise subspace and the signal subspace. The theory behind the subspace decomposition methodology is that the projection of the observable onto the noise subspace consists of only the noise and the projection of the observable onto the signal subspace consists of only the signal.
The super resolution spectrum estimation algorithms and RELAX algorithm are capable of distinguishing closely placed frequencies (sinusoids) in spectrum in presence of noise. The frequencies do not have to be harmonically related and, unlike the Digital Fourier Transform (DFT), the signal model does not introduce any artificial periodicity. For a given bandwidth, these algorithms provide significantly higher resolution than Fourier Transform. Thus, the Direct Line Of Sight (DLOS) can be reliably distinguished from other multi-paths (MP) with high accuracy. Similarly, applying the thresholded method, which will be explained later, to the artificially produced synthetic wider bandwidth ranging signal makes it possible to reliably distinguish DLOS from other paths with high accuracy.
In accordance with the exemplary embodiment, the Digital signal processing (DSP), can be employed by the multi-path mitigation processor to reliably distinguish the DLOS from other MP paths. A variety of super-resolution algorithms/techniques exist in the spectral analysis (spectrum estimation) technology. Examples include subspace based methods: MUltiple SIgnal Characterization (MUSIC) algorithm or root-MUSIC algorithm, Estimation of Signal Parameters via Rotational Invariance Techniques (ESPRIT) algorithm, Pisarenko Harmonic Decomposition (PHD) algorithm, RELAX algorithm, etc.
In all of the aforementioned super-resolution algorithms the incoming (i.e., received) signal is modeled as a linear combination of complex exponentials and their complex amplitudes of frequencies. In case of a multi-path, the received signal will be as follows:
where β×e^{i2πf×t }is the transmitted signal, f is the operating frequency, L is the number of multi-path components, and α_{K}=|α_{K}|×e^{jθ}^{K }and τ_{K }are the complex attenuation and propagation delay of the K-th path, respectively. The multi-path components are indexed so that the propagation delays are considered in ascending order. As a result, in this model τ_{0 }denotes the propagation delay of the DLOS path. Obviously, the τ_{0 }value is of the most interest, as it is the smallest value of all τ_{K}. The phase θ_{K }is normally assumed random from one measurement cycle to another with a uniform probability density function U (0,2π). Thus, we assume that α_{K}=cost (i.e., constant value)
Parameters α_{K }and τ_{K }are random time-variant functions reflecting motions of people and equipment in and around buildings. However, since the rate of their variations is very slow as compared to the measurement time interval, these parameters can be treated as time-invariant random variables within a given measurement cycle.
All these parameters are frequency-dependent since they are related to radio signal characteristics, such as, transmission and reflection coefficients. However, in the exemplary embodiment, the operating frequency changes very little. Thus, the abovementioned parameters can be assumed frequency-independent.
Equation (1) can be presented in frequency domain as:
where: A(f) is complex amplitude of the received signal, (2π×τ_{K}) are the artificial “frequencies” to be estimated by a super-resolution algorithm and the operating frequency f is the independent variable; α_{K }is the K-th path amplitude.
In the equation (2) the super-resolution estimation of (2π×τ_{K}) and subsequently τ_{K }values are based on continuous frequency. In practice, there is a finite number of measurements. Thus, the variable f will not be a continuous variable, but rather a discrete one. Accordingly, the complex amplitude A(f) can be calculated as follows:
where Â(f_{n}) are discrete complex amplitude estimates (i.e., measurements) at discrete frequencies f_{n}.
In equation (3) Â(f_{n}) can be interpreted as an amplitude and a phase of a sinusoidal signal of frequency f_{n }after it propagates through the multi-path channel. Note that all spectrum estimation based super-resolution algorithms require complex input data (i.e. complex amplitude).
In some cases, it is possible to convert real signal data, e.g. Re(Â(f_{n})), into a complex signal (e.g., analytical signal). For example, such a conversion can be accomplished by using Hilbert transformation or other methods. However, in case of short distances the value τ_{0 }is very small, which results in very low (2π×τ_{K}) “frequencies”.
These low “frequencies” create problems with Hilbert transform (or other methods) implementations. In addition, if only amplitude values (e.g., Re(Â(f_{n}))) are to be used, then the number of frequencies to be estimated will include not only the (2π×τ_{K}) “frequencies”, but also theirs combinations. As a rule, increasing the number of unknown frequencies impacts the accuracy of the super-resolution algorithms. Thus, reliable and accurate separation of DLOS path from other multi-path (MP) paths requires complex amplitude estimation.
The following is a description of a method and the multi-path mitigation processor operation during the task of obtaining complex amplitude Â(f_{n}) in presence of multi-path. Note that, while the description is focused on the half-duplex mode of operation, it can be easily extended for the full-duplex mode. The simplex mode of operation is a subset of the half-duplex mode, but would require additional events synchronization.
In half-duplex mode of operation the reader (often referred to as the “master”) and the tags (also referred to as “slaves” or “targets”) are controlled by a protocol that only allows the master or the slave to transmit at any given time. In this mode of operation the tags (target devices) serve as Transponders. The tags receive the ranging signal from a reader (master device), store it in the memory and then, after certain time (delay), re-transmit the signal back to the master.
An example of ranging signal is shown in FIG. 1 and FIG. 1A. The exemplary ranging signal employs different frequency components that are contiguous. Other waveforms, including pseudo random, spaced in frequency and/or time or orthogonal, etc. can be also used for as long as the ranging signal bandwidth remains narrow. In FIG. 1 the time duration T_{f }for every frequency component is long enough to obtain the ranging signal narrow-bandwidth property.
Another variation of a ranging signal with different frequency components is shown on FIG. 2. In includes multiple frequencies (f_{1}, f_{2}, f_{3}, f_{4}, f_{n}) transmitted over long period of time to make individual frequencies narrow-band. Such signal is more efficient, but it occupies in a wide bandwidth and a wide bandwidth ranging signal impacts the SNR, which, in turn, reduces the operating range. Also, such wide bandwidth ranging signal will violate FCC requirements on the VHF band or lower frequencies bands.
Master devices and Tag devices are identical and can operate either in Master or Transponder mode. All devices include data/remote control communication channels. The devices can exchange the information and master device(s) can remotely control tag devices. In this example depicted in FIG. 1 during an operation of a master (i.e., reader) multi-path mitigation processor originates the ranging signal to tag(s) and, after a certain delay, the master/reader receives the repeated ranging signal from the tag(s).
Thereafter, master's multi-path mitigation processor compares the received ranging signal with the one that was originally sent from the master and determines the Â(f_{n}) estimates in form of an amplitude and a phase for every frequency component f_{n}. Note that in the equation (3) Â(f_{n}) is defined for one-way ranging signal trip. In the exemplary embodiment the ranging signal makes a round-trip. In other words, it travels both ways: from a master/reader to a target/slave and from the target/slave back to the master/reader. Thus, this round-trip signal complex amplitude, which is received back by the master, can be calculated as follows:
|Â_{RT}(f_{n})|=|Â(f_{n})|^{2 }and ∠Â_{RT}(f_{n})=2×(∠Â(f_{n})) (4)
There are many techniques available for estimating the complex amplitude and phase values, including, for example, matching filtering |Â(f_{n})| and ∠Â(f_{n}). According to the exemplary embodiment, a complex amplitude determination is based on |Â(f_{n})| values derived from the master and/or tag receiver RSSI (Received Signal Strength Indicator) values. The phase values ∠Â_{RT}(f_{n}) are obtained by comparing the received by a reader/master returned base-band ranging signal phase and the original (i.e., sent by reader/master) base band ranging signal phase. In addition, because master and tag devices have independent clock systems a detailed explanation of devices operation is augmented by analysis of the clock accuracy impact on the phase estimation error. As the above description shows, the one-way amplitude |Â(f_{n})| values are directly obtainable from target/slave device. However, the one-way phase ∠Â(f_{n}) values cannot be measured directly.
In the exemplary embodiment, the ranging base band signal is the same as the one depicted in FIG. 1. However, for the sake of simplicity, it is assumed herein that the ranging base band signal consists of only two frequency components each containing multiple periods of cosine or sine waves of different frequency: F_{1 }and F_{2}. Note that F_{1}=f_{1 }and F_{2}=f_{2}. The number of periods in a first frequency component is L and the number of periods in a second frequency component is P. Note that L may or may not be equal to P, because for T_{f}=constant each frequency component can have different number of periods. Also, there is no time gap between each frequency component, and both F_{1 }and F_{2 }start from the initial phase equal to zero.
FIGS. 3A, 3B and 3C depict block diagrams of a master or a slave unit (tag) of an RF mobile tracking and locating system. F_{OSC }refers to the frequency of the device system clock (crystal oscillator 20 in FIG. 3A). All frequencies generated within the device are generated from this system clock crystal oscillator. The following definitions are used: M is a master device (unit); AM is a tag (target) device (unit). The tag device is operating in the transponder mode and is referred to as transponder (AM) unit.
In the preferred embodiment the device consists of the RF front-end and the RF back-end, base-band and the multi-path mitigation processor. The RF back-end, base-band and the multi-path mitigation processor are implemented in the FPGA 150 (see FIGS. 3B and 3C). The system clock generator 20 (see FIG. 3A) oscillates at: F_{OSC}=20 MHz; or ω_{OSC}=2π×20×10^{6}. This is an ideal frequency because in actual devices the system clocks frequencies are not always equal to 20 MHz: F_{OSC}^{M}=F_{OSC}γ^{M}; F_{OSC}^{AM}=F_{OSC}γ^{AM}.
Note that
It should be noted that other than 20 MHz F_{OSC }frequencies can be used without any impact on system performance.
Both units' (master and tag) electronic makeup is identical and the different modes of operations are software programmable. The base band ranging signal is generated in digital format by the master' FPGA 150, blocks 155-180 (see FIG. 2B). It consists of two frequency components each containing multiple periods of cosine or sine waves of different frequency. At the beginning, t=0, the FPGA 150 in a master device (FIG. 3B) outputs the digital base-band ranging signal to its up-converter 50 via I/Q DACs 120 and 125. The FPGA 150 starts with F_{1 }frequency and after time T_{1 }start generating F_{2 }frequency for time duration of T_{2}.
Since crystal oscillator's frequency might differ from 20 MHz the actual frequencies generated by the FPGA will be F_{1}γ^{M }and F_{2}γ^{M}. Also, time T_{1 }will be T_{1}β^{M }and T_{2 }will be T_{2}β^{M}. It is also assumed that T_{1},T_{2},F_{1},F_{2 }are such that F_{1}γ^{M}*T_{1}β^{M}=F_{1}T_{1 }and F_{2}γ^{M}*T_{2}β^{M}=F_{2}T_{2}, where both F_{1}T_{1 }& F_{2}T_{2 }are integer numbers. That means that the initial phases of F_{1 }and F_{2 }are equal to zero.
Since all frequencies are generated from the system crystal oscillator 20 clocks, the master' base-band I/Q DAC(s) 120 and 125 outputs are as follows: F_{1}=γ^{M}20×10^{6}×K_{F}_{1 }and F_{2}=γ^{M}20×10^{6}×K_{F}_{2}, where K_{F}_{1 }and K_{F}_{2 }are constant coefficients. Similarly, the output frequencies TX_LO and RX_LO from frequency synthesizer 25 (LO signals for mixers 50 and 85) can be expressed through constant coefficients. These constant coefficients are the same for the master (M) and the transponder (AM)—the difference is in the system crystal oscillator 20 clock frequency of each device.
The master (M) and the transponder (AM) work in a half-duplex mode. Master' RF front-end up-converts the base-band ranging signal, generated by the multi-path mitigation processor, using quadrature up-converter (i.e., mixer) 50 and transmits this up-converted signal. After the base-band signal is transmitted the master switches from TX to RX mode using RF Front-end TX/RX Switch 15. The transponder receives and down-converts the received signal back using its RF Front-end mixer 85 (producing First IF) and ADC 140 (producing Second IF).
Thereafter, this second IF signal is digitally filtered in the Transponder RF back-end processor using digital filters 190 and further down-converted to the base-band ranging signal using the RF back-end quadrature mixer 200, digital I/Q filters 210 and 230, a digital quadrature oscillator 220 and a summer 270. This base-band ranging signal is stored in the transponder's memory 170 using Ram Data Bus Controller 195 and control logic 180.
Subsequently, the transponder switches from RX to TX mode using RF front-end switch 15 and after certain delay t_{RTX }begins re-transmitting the stored base-band signal. Note that the delay is measured in the AM (transponder) system clock. Thus, t_{RTX}^{AM}=t_{RTX}β^{AM}. The master receives the transponder transmission and down-converts the received signal back to the base-band signal using its RF back-end quadrature mixer 200, the digital I and Q filters 210 and 230, the digital quadrature oscillator 220 (see FIG. 3C).
Thereafter, the master calculates the phase difference between F_{1 }and F_{2 }in the received (i.e., recovered) base-band signal using multi-path mitigation processor arctan block 250 and phase compare block 255. The amplitude values are derived from the RF back-end RSSI block 240.
For improving the estimation accuracy it is always desirable to improve the SNR of the amplitude estimates from block 240 and phase difference estimates from block 255. In the preferred embodiment the multi-path mitigation processor calculates amplitude and phase difference estimates for many time instances over the ranging signal frequency component duration (T_{f}). These values, when averaged, improve SNR. The SNR improvement can be in an order that is proportional to √{square root over (N)}, where N is a number of instances when amplitude and phase difference values were taken (i.e., determined).
Another approach to the SNR improvement is to determine amplitude and phase difference values by applying matching filter techniques over a period of time. Yet, another approach would be to estimate the phase and the amplitude of the received (i.e., repeated) base band ranging signal frequency components by sampling them and integrating over period T≦T_{f }against the original (i.e., sent by the master/reader) base-band ranging signal frequency components in the I/Q form. The integration has the effect of averaging of multiple instances of the amplitude and the phase in the I/Q format. Thereafter, the phase and the amplitude values can be translated from the I/Q format to the |Â(f_{n})| and ∠Â(f_{n}) format.
Let's assume that t=0 under master' multi-path processor control the master base-band processor (both in FPGA 150) start the base-band ranging sequence.
φ_{FPGA}^{M}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(t)), t<T_{1}β^{M}, t<T_{1}β^{M};
φ_{FPGA}^{M}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(T_{1}β^{M})+K_{F}_{2}(t−T_{1}β^{M})), t>T_{1}β^{M},
where T_{f}≧T_{1}β^{M}.
The phase at master's DAC(s) 120 and 125 outputs are as follows:
φ_{DAC}^{M}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(t−t_{DAC}^{M}))+φ_{DAC}^{M}(0), t<T_{1}β^{M}+t_{DAC}^{M};
φ_{DAC}^{M}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(T_{1}β^{M})+K_{F}_{2}(t−T_{1}β^{M}−t_{DAC}^{M}))+φ_{DAC}^{M}(0), t>T_{1}β^{M}+t_{DAC}^{M}
Note that DACs 120 and 125 have internal propagation delay, t_{DAC}^{M}, that does not depend upon the system clock.
Similarly, the transmitter circuitry components 15, 30, 40 and 50 will introduce additional delay, t_{TX}^{M}, that does not depend upon the system clock.
As a result, the phase of the transmitted RF signal by the master can be calculated as follows:
φ_{RF}^{M}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(t−t_{DAC}^{M}−t_{TX}^{M})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}))+φ_{DAC}^{M}(0)+φ_{SYN}_{—}_{TX}^{M}(0),
t<T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M};
φ_{RF}^{M}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(T_{1}β^{M})+K_{F}_{2}(t−T_{1}β^{M}−t_{DAC}−t_{TX}^{M})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}))+φ_{DAC}^{M}(0)+φ_{SYN}_{—}_{TX}^{M}(0),
t>T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M}
The RF signal from the master (M) experiences a phase shift φ^{MULT }that is a function of the multi-path phenomena between the master and tag.
The φ^{MULT }values depend upon the transmitted frequencies, e.g. F_{1 }and F_{2}. The transponder (AM) receiver' is not able to resolve each path because of limited (i.e., narrow) bandwidth of the RF portion of the receiver. Thus, after a certain time, for example, 1 microsecond (equivalent to ˜300 meters of flight), when all reflected signals have arrived at the receiver antenna, the following formulas apply:
φ_{ANT}^{AM}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(t−t_{DAC}^{M}−t_{TX}^{M})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}))+φ_{F}_{1}^{MULT}+φ_{DAC}^{M}(0)+φ_{SYN}_{—}_{TX}^{M}(0),
10^{−6}<t<T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M};
φ_{ANT}^{AM}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(T_{1}β^{M})+K_{F}_{2}(t−T_{1}β^{M}−t_{DAC}^{M}−t_{TX}^{M})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}))+φ_{F}_{2}^{MULT}+φ_{DAC}^{M}(0)+φ_{SYN}_{—}_{TX}^{M}(0),
t>T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M}+10^{−6}
In the AM (transponder) receiver at the first down converter, element 85, an output, e.g. first IF, the phase of the signal is as follows:
φ_{IF}_{—}_{1}^{AM}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(t−t_{DAC}^{M}−t_{TX}^{M}−t_{RX}^{AM})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}−t_{RX}^{AM}))−
γ^{AM}×ω_{OSC}×(K_{SYN}_{—}_{RX}_{—}_{1}t))+
φ_{F}_{1}^{MULT}+φ_{SYN}_{—}_{TX}^{M}(0)−φ_{SYN}_{—}_{RX}_{—}_{1}^{AM}(0), 10^{−6}<t<T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M}+t_{RX}^{AM};
φ_{IF}_{—}_{1}^{AM}(t)=γ^{M}×ω_{OSC}×(K_{F}_{1}(T_{1}β^{M})+K_{F}_{2}(t−T_{1}β^{M}−t_{DAC}^{M}−t_{TX}^{M}−t_{RX}^{AM})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}−t_{RX}^{AM}))
−γ^{AM}×ω_{OSC}×(K_{SYN}_{—}_{RX}_{—}_{1}t)+
φ_{F2}^{MULT}+φ_{SYN}_{—}_{TX}^{M}(0)−φ_{SYN}_{—}_{RX}_{—}_{1}^{AM}(0), t>T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M}+t_{RX}^{AM}+10^{−6}
Note that the propagation delay t_{RX}^{AM }in the receiver RF section (elements 15 and 60-85) does not depend upon the system clock. After passing through RF Front-end filters and amplifiers (elements 95-110 and 125) the first IF signal is sampled by the RF Back-end ADC 140. It is assumed that ADC 140 is under-sampling the input signal (e.g., first IF). Thus, the ADC also acts like a down-converter producing the second IF. The first IF filters, amplifiers and the ADC add propagation delay time. At the ADC output (second IF):
φ_{ADC}^{AM}(t)=
γ^{M}×ω_{OSC}×(K_{F}_{1}(t−t_{DAC}^{M}−t_{TX}^{M}−t_{RX}^{AM}−t_{IF}_{—}_{1}^{AM}−t_{ADC}^{AM})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}−t_{RX}^{AM}−t_{IF}_{—}_{1}^{AM}−t_{ADC}^{AM}))−
γ^{AM}×ω_{OSC}×(K_{SYN}_{—}_{RX}_{—}_{1}(t−t_{IF}_{—}_{1}^{AM}−t_{ADC}^{AM})+K_{ADC}(t))+φ_{F}_{1}^{MULT}+φ_{SYN}_{—}_{TX}^{M}(0)−φ_{SYN}_{—}_{RX}_{—}_{1}^{AM}(0)−φ_{ADC}_{—}_{CLK}^{AM}(0),
10^{−6}<t<T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M}+t_{RX}^{AM}+t_{IF}_{—}_{1}^{AM}+t_{ADC}^{AM};
φ_{ADC}^{AM}(t)=
γ^{M}×ω_{OSC}×(K_{F}_{1}(T_{1}β^{M})+K_{F}_{2}(t−T_{1}β^{M}−t_{DAC}^{M}−t_{TX}^{M}−t_{RX}^{AM}−t_{IF}_{—}_{1}^{AM}−t_{ADC}^{AM})+K_{SYN}_{—}_{TX}(t−t_{TX}^{M}−t_{RX}^{AM}−t_{IF}_{—}_{1}^{AM}−t_{ADC}^{AM}))
−γ^{AM}×ω_{OSC}×(K_{SYN}_{—}_{RX}_{—}_{1}(t−t_{IF}_{—}_{1}^{AM}−t_{ADC}^{AM})+K_{ADC}(t))+φ_{F2}^{MULT}+φ_{SYN}_{—}_{TX}^{M}(0)−φ_{SYN}_{—}_{RX}_{—}_{1}^{AM}(0)−φ_{ADC}_{—}_{CLK}^{AM}(0),
t>T_{1}β^{M}+t_{DAC}^{M}+t_{TX}^{M}+t_{RX}^{AM}+t_{IF}_{—}_{1}^{AM}+t_{ADC}^{AM}+10^{−6}
In the FPGA 150 the second IF signal (from the ADC output) is filtered by the RF Back-end digital filters 190 and further down-converted back to base-band ranging signal by the third down-converter (i.e., quadrature mixer 200, digital filters 230 and 210 and digital quadrature oscillator 220), summed in the summer 270 and is stored in the memory 170. At the third down-converter output (i.e., quadrature mixer):
Note that propagation delay t_{FIR}^{AM}=t_{FIR}β^{AM }in the FIR section 190 does not depend upon the system clock.
After RX→TX delay the stored (in memory 170) base-band ranging signal from the master (M) is retransmitted. Note that RX→TX delay t_{RTX}^{AM}=t_{RTX}β^{AM}.
By the time the signal from the transponder reaches the master' (M) receiver antenna the RF signal from transponder (AM) experiences another phase shift φ^{MULT }that is a function of the multi-path. As discussed above, this phase shift happens after a certain time period when all reflected signals have arrived at the master' receiver antenna:
In the master receiver the signal from transponder goes through the same down-conversion process as in the transponder receiver. The result is the recovered base-band ranging signal that was originally sent by the master.
For the first frequency component F_{1}:
For the second frequency component F_{2}:
Substitutions:
T_{D}_{—}_{M-AM}=t_{DAC}^{M}+t_{TX}^{M}+t_{RX}^{AM}+t_{IF}_{—}_{1}^{AM}+t_{ADC}^{AM}+t_{FIR}β^{AM}+t_{RTX}β^{AM}+t_{DAC}^{AM}+t_{TX}^{AM}+t_{RX}^{M}+t_{IF}_{—}_{1}^{M}+t_{ADC}^{M}+t_{FIR}β^{M};
where T_{D}_{—}_{M-AM }is the propagation delay through master (M) and transponder (AM) circuitry.
φ_{BB}_{—}_{M-AM}(0)=φ_{SYN}_{—}_{TX}^{M}(0)−φ_{SYN}_{—}_{RX}_{—}_{1}^{AM}(0)−
φ_{ADC}_{—}_{CLK}^{AM}(0)−φ_{SYN}_{—}_{RX}_{−}_{2}^{AM}(0)+φ_{SYN}_{—}_{TX}^{AM}−φ_{SYN}_{—}_{RX}_{—1}^{M}(0)−φ_{ADC}_{—}_{CLK}^{M}(0)−φ_{SYN}_{—}_{RX}_{—}_{2}^{M}(0)=Const;
where: φ_{BB}_{—}_{M-AM}(0) is the LO phase shift, at time t=0, from master (M) and transponder (AM) frequency mixers, including ADC(s).
Also:
K_{SYN}_{—}_{TX}=K_{SYN}_{—}_{RX}_{—}_{1}+K_{ADC}+K_{SYN}_{—}_{RX}_{—}_{2}
First frequency component F_{1}:
First frequency component F_{1 }continued:
Second frequency component F_{2}:
Second frequency component F_{2}, continued:
Further substituting:
where α is a constant.
Then the final phase equations is:
φ_{BB}_{—}_{RECOV}^{M}(t)=
γ^{M}×ω_{OSC}×(K_{F}_{1}(t−T_{D}_{—}_{M-AM}))+2×φ_{F}_{1}^{MULT}+φ_{BB}_{—}_{M-AMN}(0)+α,
2×10^{−6}<t<T_{1}β^{M}+T_{D}_{—}_{M-AM};
φ_{BB}_{—}_{RECOV}^{M}(t)=
γ^{M}×ω_{OSC}×(K_{F}_{1}(T_{1}β^{M})+K_{F}_{2}(t−T_{1}β^{M}−T_{D}_{—}_{M-AM}))+2×φ_{F}_{2}^{MULT}+φ_{BB}_{—}_{M-AM}(0)+α,
t>T_{1}β^{M}+T_{D}_{—}_{M-AM}+2×10^{−6} (5)
From the equation (5):
where i=2, 3, 4 . . . ; and 2×ΔΦ_{F}_{1}_{/F}_{i }is equal to 2×(φ_{F}_{i}^{MULT}−φ_{F}_{1}^{MULT}).
For example, the difference 2×(φ_{F}_{2}^{MULT}−φ_{F}_{1}^{MULT}) at time instances t1 and t2:
2×φ_{F}_{2}^{MULT}−2×φ_{F}_{1}^{MULT}=2×ΔΦ_{F}_{1}_{/F}_{2}=
φ_{BB}_{—}_{RECOV}^{M}(t_{2})−φ_{BB}_{—}_{RECOV}^{M}(t_{1})−γ^{M}×ω_{OSC}×[K_{F}_{1}(T_{1}β^{M})+(K_{F}_{2}(t_{2}−T_{1}β^{M}−T_{D}_{—}_{M-AM}))−(K_{F}_{1}(t_{1}−T_{D}_{—}_{M-AM}))],
2×10^{−6}<t_{1}<T_{1}β^{M}+T_{D}_{—}_{M-AM}; t_{2}>T_{1}β^{M}+T_{D}_{—}_{M-AM}+2×10^{−6}
To find 2×ΔΦ_{F}_{1}_{/F}_{2 }difference we need to know T_{D}_{—}_{M-AM}:
T_{D}_{—}_{M-AM}=T_{LB}_{—}_{M}β^{M}+T_{LB}_{—}_{AM}β^{AM}+t_{RTX}β^{AM};
T_{LB}_{—}_{M}=t_{DAC}^{M}+t_{TX}^{M}+t_{RX}^{M}+t_{IF}_{—}_{1}^{M}+t_{ADC}^{M}+t_{FIR}β^{M}; T_{LB}_{—}_{AM}=t_{DAC}^{AM}+t_{TX}^{AM}t_{RX}^{AM}+t_{IF}_{—}_{1}^{AM}+t_{ADC}^{AM}+t_{FIR}β^{AM},
where T_{LB}_{—}_{M }and T_{LB}_{—}_{AM }are propagation delays through the master (M) and transponder (AM) TX and RX circuitries that are measured by placing devices in the loop-back mode. Note that the master and the transponder devices can measure T_{LB}_{—}_{M }and T_{LB}_{—}_{AM }automatically; and we also know the t_{RTX }value.
From the above formulas and t_{RTX }value T_{D}_{—}_{M-AM }can be determined and consequently, for a given t_{1}, and t_{2 }the 2×ΔΦ_{F}_{1}_{/F}_{2 }value can be found as follows:
Or, assuming that β^{M}=β^{AM}=1:
2×ΔΦ_{F}_{1}_{/F}_{2}=φ_{BB}_{—}_{RECOV}^{M}(t_{2})−φ_{BB}_{—}_{RECOV}^{M}(t_{1})−
γ^{M}×ω_{OSC}×[K_{F}_{2}t_{2}−K_{F}_{1}t_{1}−(K_{F}_{2}−K_{F}_{1})×(T_{1}−T_{D}_{—}_{M-AM})],
2×10^{−6}<t_{1}<T_{1}+T_{D}_{—}_{M-AM}; t_{2}=t_{1}+T_{1}; (6A)
From the equation (6) it can be concluded that at operating frequency(s) ranging signal(s) complex amplitude values can be found from processing the returned base-band ranging signal.
The initial phase value 2×φ_{F}_{1}^{MULT }can be assumed to be equal zero because the subspace algorithms are not sensitive to a constant phase offset. If necessary, the 2×φ_{F}_{1}^{MULT }value (phase initial value) can be found by determining the TOA (Time Of Arrival) using the narrow-bandwidth ranging signal method as described in a co-pending application Ser. No. 11/670,595, incorporated herein by reference in its entirety. This method estimates the ranging signal round trip delay, which is equal to 2×T_{FLT}β^{M }and the 2×φ_{F}_{1}^{MULT }value can be found from the following equation:
2×φ_{F}_{1}^{MULT}=2×β^{M}×γ^{M}×ω_{OSC}×(K_{SYN}_{—}_{TX}+K_{F}_{1})×(T_{FLT}),
Or:
2×φ_{F}_{1}^{MULT}=2×ω_{OSC}×(K_{SYN}_{—}_{TX}+K_{F}_{1})×(T_{FLT}),
In the preferred embodiment, the returned base-band ranging signal phase values φ_{BB}_{—}_{RECOV}^{M}(t) are calculated by the multi-path processor' arctan block 250. To improve SNR, the multi-path mitigation processor phase compare block 255 calculates 2×ΔΦ_{F}_{1}_{/F}_{i}=φ_{BB}_{—}_{RECOV}^{M}(t_{m})−φ_{BB}_{—}_{RECOV}^{M}(t_{n}) for many instances n (n=2, 3, 4 . . . ) using the equation (6A), and then average them out to improve SNR. Note that 2×10^{−6}<t_{n}<T_{f}+T_{D}_{—}_{M-AM}; t_{m}=t_{1}+T_{f}.
From the equations 5 and 6 it becomes apparent that the recovered (i.e., received) base-band ranging signal has the same frequency as the original base-band signal that as sent by the master. Thus, there is no frequency translation despite the fact that the master (M) and the transponder (AM) system clocks can differ. Because the base-band signal consists of several frequency components, each component is consists of multiple periods of a sinusoid, it is also possible to estimate the phase and the amplitude of the received ranging signal by sampling the received base-band signal individual component frequency with the corresponding original (i.e., sent by the master) base-band signal individual frequency component and integrating the resulting signal over period T≦T_{f}.
This operation generates complex amplitude values Â_{RT }(f_{n}) of received ranging signal in the I/Q format. Note that each base-band signal individual frequency component that was sent by the master has to be shifted in time by the T_{D}_{—}_{M-AM}. The integration operation produces effect of averaging out the multiple instances of the amplitude and the phase (e.g., increasing the SNR). Note that the phase and the amplitude values can be translated from the I/Q format to the |Â(f_{n})| and ∠Â_{RT}(f_{n}) format.
This method of sampling, integrating over period of T≦T_{f }and subsequent conversion from the I/Q format to the |Â(f_{n})| and ∠Â(f_{n}) format can be implemented in the phase compare block 255 in FIG. 3C. Thus, depending upon the block's 255 design and implementation, either the method of the preferred embodiment, based on the equation (5), or an alternative method, described in this section, can be used.
Although the ranging signal bandwidth is narrow, the frequency difference f_{n}−f_{1 }can be relatively large, for example, in an order of several megahertz. As a result, the receiver's bandwidth has to be kept wide enough to pass all of the f_{1}:f_{n }ranging signal frequencies components. This wide receiver bandwidth impacts the SNR. To reduce the receiver effective bandwidth and improve the SNR, the received ranging signal base-band frequency components can be filtered by the RF back-end processor in FPGA 150 by the digital narrow bandwidth filters tuned for each individual frequency component of the received base-band ranging signal. However, this large number of digital filters (the number of filters equals to the number of individual frequency components, n) puts additional burden on the FPGA resources, increasing its cost, size and power consumption.
In the preferred embodiment only two narrow bandwidth digital filters are used: one filter is always tuned for f_{1 }frequency component and the other filter can be tuned for all other frequencies components: f_{2}:f_{n}. Multiple instances of ranging signal are sent by the master. Each instance consists of only two frequencies: f_{1}:f_{2}; f_{1}:f_{3 }. . . ; f_{1}:f_{i }. . . ; f_{1}:f_{n}. Similar strategies are also possible.
Please note that it is also entirely possible to keep the base-band ranging signal components to only two (or even one) generating the rest of the frequency components by adjusting the frequency synthesizers, e.g. changing K_{SYN}. It is desirable that LO signals for up-converters and down-converters mixers are generated using the Direct Digital Synthesis (DDS) technology. For high VHF band frequencies this can present an undesired burden on the transceiver/FPGA hardware. However, for lower frequencies this might be a useful approach. Analog frequency synthesizers can also be used, but may take additional time to settle after frequency is changed. Also, in case of analog synthesizers, two measurements at the same frequency would have to be made in order to cancel a phase offset that might develop after changing the analog synthesizer's frequency.
The actual T_{D}_{—}_{M-AM }that is used in the above equations is measured in both: the master (M) and the transponder (AM) systems clocks, e.g. T_{LB}_{—}_{AM }and t_{RTX }are counted in the transponder (AM) clocks and T_{LB}_{—}_{M }is counted in the master (M) clock. However, when 2×ΔΦ_{F}_{1}_{/F}_{2 }is calculated both: T_{LB}_{—}_{AM }and t_{RTX }are measured (counted) in master (M) clock. This introduces an error:
2×ΔΦ_{ERROR}=γ^{M}×ω_{OSC}×(K_{F}_{2}−K_{F}_{1})×(T_{LB}_{—}_{AM}(β^{AM}β^{M}−β_{AM})+t_{RTX}(β^{M}−β^{AM})) (7)
The phase estimation error (7) impacts the accuracy. Therefore, it is necessary to minimize this error. If β^{M}=β^{AM}, in other words, all master(s) and transponders (tags) system clocks are synchronized, then the contribution from the t_{RTX }time is eliminated.
In the preferred embodiment, the master and the transponder units (devices) are capable of synchronizing clocks with any of the devices. For example, a master device can serve as a reference. Clock synchronization is accomplished by using the remote control communication channel, whereby under FPGA 150 control, the frequency of temperature compensated crystal oscillator TCXO 20 is adjusted. The frequency difference is measured at the output of the summer 270 of the master device while the selected transponder device is transmitting a carrier signal.
Thereafter, the master sends a command to the transponder to increase/decrease TCXO frequency. This procedure may be repeated several times to achieve greater accuracy by minimizing frequency at the summer 270 output. Please note that in an ideal case the frequency at the summer 270 output should become equal to zero. An alternative method is to measure the frequency difference and make a correction of the estimated phase without adjusting the transponder' TCXO frequency.
While β^{M}−β^{AM }can be considerably reduced there is a phase estimation error when β^{M}≠1. In this case the margin of error depends upon a long term stability of the reference device (usually master (M)) clock generator. In addition, the process of clock synchronization may take considerable amount of time, especially with large number of units in the field. During the synchronization process the track-locate system becomes partially or fully inoperable, which negatively impacts the system readiness and performance. In this case the abovementioned method that does not require the transponder' TCXO frequency adjustment is preferred.
Commercially available (off the shelf) TCXO components have high degree of accuracy and stability. Specifically, TCXO components for the GPS commercial applications are very accurate. With these devices, the phase error impact on locating accuracy can be less than one meter without the need for frequent clock synchronization.
After narrow bandwidth ranging signal multi-path mitigation processor obtains the returned narrow bandwidth ranging signal complex amplitude Â_{RT}(f_{n}), the further processing (i.e., execution of super-resolution algorithms), is implemented in the software-based component, which is a part of the multi-path mitigation processor. This software component can be implemented in the master (reader) host computer CPU and/or the microprocessor that is embedded in the FPGA 150 (not shown). in the preferred embodiment the multi-path mitigation algorithm(s) software component is executed by the master host computer CPU.
The super-resolution algorithm(s) produce estimation of (2π×τ_{K}) “frequencies”, e.g. τ_{K }values. At the final step the multi-path mitigation processor selects τ with the smallest value (i.e., the DLOS delay time).
In certain cases where the ranging signal narrow bandwidth requirements are somewhat relaxed, the DLOS path can be separated from MP paths by employing a continuous (in time) chirp. In the preferred embodiment this continuous chirp is Linear Frequency Modulation (LFM). However, other chirp waveforms can be also used.
Let's assume that under multi-path mitigation processor control a chirp with bandwidth of B and duration of T is transmitted. That gives a chirp rate of
radians per second. Multiple chirps are transmitted and received back. Note that chirps signals are generated digitally with each chirp started at the same phase.
In the multi-path processor each received single chirp is aligned so that the returned chirp is from the middle of the area of interest.
The chirp waveform equation is:
s(t)=exp(i(ω_{0}t+βt^{2})),
where ω_{0 }is the initial frequency for 0<t<T.
For a single delay round-trip τ, e.g. no multi-path, the returned signal (chirp) is s(t−τ).
The multi-path mitigation processor then “deramps” the s(t−τ) by performing complex conjugate mix with the originally transmitted chirp. The resulting signal is a complex sinusoid:
f_{τ}(t)=exp(−ω_{0}τ)exp(−2iβτt)exp(iβτ^{2}), (8)
where exp(−iw_{0}τ_{k}) is the amplitude and 2βτ is the frequency and 0≦t≦T. Note that the last term is a phase and it is negligible.
In case of multi-path, the composite deramped signal consists of multiple complex sinusoids:
where L is the number of ranging signal paths, including the DLOS path and 0≦t≦T.
Multiple chirps are transmitted and processed. Each chirp is individually treated/processed as described above. Thereafter, the multi-path mitigation processor assembles results of individual chirps processing:
where N is the number of chirps,
is the dead time zone between two consecutive chirps; 2βτ_{k }are artificial delay “frequencies”. Again, the most interesting is the lowest “frequency”, which corresponds to the DLOS path delay.
In the equation (10) f_{MP}^{N}(t) can be thought of as N samples of a sum of complex sinusoids at times:
0≦t_{α}≦T; t_{1}=t_{α}+ρ; t_{2}=t_{α}+2ρ . . . ; t_{m−1}=t_{α}+(N−1)ρ; mε0:m−1;
Thus, the number of samples can be a multiple of N, e.g. αN; α=1, 2, . . . .
From the equation (10) the multi-path mitigation processor produces αN complex amplitude samples in time domain that are used in further processing (i.e., execution of super-resolution algorithms). This further processing is implemented in the software component, which is a part of the multi-path mitigation processor. This software component can be executed by the master (reader) host computer CPU and/or by the microprocessor that is embedded in the FPGA 150 (not shown), or both. In the preferred embodiment the multi-path mitigation algorithm(s) software is executed by the master host computer CPU.
The super-resolution algorthm(s) produce estimation of 2βτ_{k }“frequencies”, e.g. τ_{K }values. At the final step the multi-path mitigation processor selects τ with the smallest value, i.e. the DLOS delay time.
An explanation will be given of a special processing method, called the “threshold technique,” which can serve as an alternative to the super-resolution algorithms. In other words, it is used to enhance reliability and accuracy in distinguishing DLOS path from other MP paths using the artificially generated synthetic wider bandwidth ranging signal.
The frequency domain base-band ranging signal shown in FIG. 1 and FIG. 1A can be converted into time domain base-band signal s(t):
It is readily verified that s(t) is periodic with period 1/Δt, and for any integer k, that s(k/Δt)=2N+1, which is the peak value of the signal. Where n=N in FIG. 1 and FIG. 1A.
FIG. 4 shows two periods of s(t) for the case where N=11 and Δf=250 kHz. The signal appears as a sequence of pulses of height 2N+1=23 separated by 1/Δf=4 microseconds. Between the pulses is a sinusoidal waveform with varying amplitude and 2N zeros. The wide bandwidth of the signal can be attributed to the narrowness of the tall pulses. It can be also seen that the bandwidth extends from zero frequency to NΔf=2.75 MHz.
The basic idea of the thresholded method that is used in the preferred embodiment is to enhance the artificially generated synthetic wider bandwidth ranging reliability and accuracy in distinguishing DLOS path from other MP paths. The threshold method detects when the start of the leading edge of a wideband pulse arrives at a receiver. Because of filtering in the transmitter and receiver, the leading edge does not rise instantaneously, but rises out of the noise with smoothly increasing slope. The TOA of the leading edge is measured by detecting when the leading edge crosses a predetermined threshold T.
A small threshold is desirable because it gets crossed sooner and the error delay τ between the true start of the pulse and the threshold crossing is small. Thus, any pulse replica arriving due to multi-path has no effect if the start of the replica having a delay greater than τ. However, the presence of noise places a limit on how small the threshold T can be. One way to decrease the delay τ is to use the derivative of the received pulse instead of the pulse itself, because the derivative rises faster. The second derivative has an even faster rise. Higher order derivatives might be used, but in practice they can raise the noise level to an unacceptable value, so the thresholded second derivative is used.
Although the 2.75 MHz wide signal depicted in FIG. 4 has a fairly wide bandwidth, it is not suitable for measuring range by the abovementioned method. That method requires transmitted pulses each having a zero-signal precursor. However, it is possible to achieve that goal by modifying the signal so that the sinusoidal waveform between the pulses is essentially cancelled out. In the preferred embodiment it is done by constructing a waveform which closely approximates the signal on a chosen interval between the tall pulses, and then subtracting it from the original signal.
The technique can be illustrated by applying it to the signal in FIG. 1. The two black dots shown on the waveform are the endpoints of an interval I centered between the first two pulses. The left and right endpoints of the interval I, which have been experimentally determined to provide the best results, are respectively at:
An attempt to generate a function g(t) which essentially cancels out the signal s(t) on this interval, but does not cause much harm outside the interval, is performed. Since the expression (11) indicates that s(t) is the sinusoid sinπ(2N+1)Δft modulated by 1/sinπΔft, first a function h(t) which closely approximates 1/sinπΔft on the interval I is found, and then form g(t) as the product:
g(t)=h(t)sinπ(2N+1)Δft (13)
h(t) is generated by the following sum:
where
φ_{0}(t)≡1, φ_{k}(t)=sin kπΔft for k=1, 2, . . . , M (15)
and the coefficients α_{k }are chosen to minimize the least-square error
over the interval I.
The solution is readily obtained by taking partial derivatives of J with respect to the α_{k }and setting them equal to zero. The result is the linear system of M+1 equations
that can be solved for the α_{k}, where
Then,
Using the definition of the functions φ_{k}(t) given by (12)
The g(t) is subtracted from s(t) to get a function r(t), which should essentially cancel s(t) on the interval I. As indicated in the Appendix, an appropriate choice for the upper limit M for the summation in the equation (20) is M=2N+1. Using this value and the results from the Appendix,
From the equation (17) it is seen that a total of 2N+3 frequencies (including the zero-frequency DC term) are required to obtain the desired signal r(t). FIG. 5 shows the resulting signal r(t) for the original signal s(t) shown in FIG. 1, where N=11. In this case the construction of r(t) requires 25 carriers (including the DC term b_{0}).
The important characteristics of r(t) as constructed above are as follows:
1. The lowest frequency is zero Hz and the the highest frequency is (2N+1)Δf Hz, as seen from (14). Thus, the total bandwidth is (2N+1)Δf Hz.
2. All carriers are cosine functions (including DC) spaced Δf apart, except for one carrier, which is a sine function located at frequency
3. Although the original signal s(t) has period 1/Δf, r(t) has period 2/Δf. The first half of each period of r(t), which is a full period of s(t), contains a cancelled portion of the signal, and the second half-period of r(t) is a large oscillatory segment. Thus, cancellation of the precursor occurs only in every other period of s(t).
This occurs because the canceling function g(t) actually strengthens s(t) in every other period of s(t). The reason is that g(t) reverses its polarity at every peak of s(t), whereas s(t) does not. A method of making every period of s(t) contain a cancelled portion to increase processing gain by 3 dB is described below.
4. The length of the cancelled portion of s(t) is about 80-90% of 1/Δf. Therefore, Δf needs to be small enough to make this length long enough to eliminate any residual signal from previous non-zero portions of r(t) due to multi-path.
5. Immediately following each zero portion of r(t) is the first cycle of an oscillatory portion. In the preferred embodiment, in the TOA measurement method as described above, the first half of this cycle is used for measuring TOA, specifically the beginning of its rise. It is interesting to note that the peak value of this first half-cycle (which will be called the main peak) is somewhat larger than the corresponding peak of s(t) located at approximately the same point in time. The width of the first half-cycle is roughly inversely proportional to NΔf.
6. A large amount of processing gain can be achieved by:
(a) Using the repetitions of the signal r(t), because r(t) is periodic with period 2/Δf. Also, an additional 3 dB of processing gain is possible by a method to be described later.
(b) Narrowband filtering. Because each of the 2N+3 carriers is a narrowband signal, the occupied bandwidth of the signal is much smaller than that of a wideband signal spread out across the entire allocated band of frequencies.
For the signal r(t) shown in FIG. 5, where N=11 and Δf=250 kHz, the length of the cancelled portion of s(t) is about 3.7 microseconds or 1,110 meters. This is more than enough to eliminate any residual signal from previous non-zero portions of r(t) due to the multi-path. The main peak has value of approximately 35, and the largest magnitude in the precursor (i.e., cancellation) region is about 0.02, which is 65 dB below the main peak. This is desirable for getting good performance using the TOA measurement thresholded technique as described above.
Use of fewer carriers is depicted in FIG. 6, which illustrates a signal that is generated using Δf=850 kHz, N=3, and M=2N+1=7, for a total of only 2N+3=9 carriers. In this case, the period of the signal is only 2/Δf≅2.35 microseconds as compared to the signal in FIG. 5, where the period is 8 microseconds. Since this example has more periods per unit time, one might expect that more processing gain could be achieved.
However, since fewer carriers are used, the amplitude of the main peak is about ⅓ as large as before, which tends to cancel the expected extra processing gain. Also, the length of the zero-signal precursor segments is shorter, about 0.8 microseconds or 240 meters. This should still be enough to eliminate any residual signal from previous non-zero portions of r(t) due to the multi-path. Note that the total bandwidth of (2N+1)Δf=5.95 MHZ is about the same as before, and that the width of the half-cycle of the main peak is also roughly the same. Since fewer carriers are used, there should be some extra processing gain when each carrier is narrowband filtered at the receiver. Moreover, the largest magnitude in the precursor (i.e., cancellation) region is now about 75 dB below the main peak, a 10 dB improvement from the previous example.
Transmission at RF Frequencies: up to this point r(t) has been described as a base-band signal for purposes of simplicity. However, it can be translated up to RF, transmitted, received, and then reconstituted as a base-band signal at the receiver. To illustrate, consider what happens to one of the frequency components ω_{k }in the base-band signal r(t) traveling via one of the multi-path propagation paths having index j (radian/sec frequencies are used for notational simplicity):
b_{k }cos ω_{k}t (at baseband in transmitter)
b_{k }cos(ω+ω_{k})t (translated by frequency ω up to RF)
a_{j}b_{k }cos[(ω+ω_{k})(t−τ_{j})+φ_{j}] (at receiver antenna)
a_{j}b_{k }cos[ω_{k}(t−τ_{j})+φ_{j}+θ] (translated by frequency−ω to baseband) (23)
It is assumed here that the transmitter and receiver are frequency synchronized. The parameter b_{k }is the k^{th }coefficient in expression (21) for r(t). The parameters τ_{j }and φ_{j }are respectively the path delay and phase shift (due to dielectric properties of a reflector) of the j^{th }propagation path. The parameter θ is the phase shift occurring in the down-conversion to base-band in the receiver. A similar sequence of functions can be presented for the sine component of the equation (21).
It is important to note that as long as the zero-signal precursors in r(t) have length sufficiently larger than the largest significant propagation delay, the final base-band signal in the equation (20) will still have zero-signal precursors. Of course, when all frequency components (index k) over all paths (index j) are combined, the base-band signal at the receiver will be a distorted version of r(t), including all phase shifts.
Sequential Carrier Transmissions and Signal Reconstruction are illustrated in FIG. 1 and FIG. 1A. It is assumed that the transmitter and the receiver are time and frequency synchronized, the 2N+3 transmitted carriers need not be transmitted simultaneously. As an example, consider the transmission of the signal whose base-band representation is that of FIG. 1A and FIG. 6.
In FIG. 6 N=3, and suppose each of the 9 frequency components for 1 millisecond are sequentially transmitted. The start and the end times for each frequency transmission are known at the receiver, so it can sequentially start and end its reception of each frequency component at those respective times. Since the signal propagation time is very short compared to 1 millisecond (it will normally be less than several microseconds in the intended application), a small portion of each received frequency component should be ignored, and the receiver can easily blank it out.
The entire process of receiving 9 frequency components can be repeated in 9-millisecond blocks of additional reception to increase the processing gain. In one second of total reception time there would be about 111 such 9-millisecond blocks available for processing gain. Additionally, within each block there would be additional processing gain available from 0.009/(2/Δf)≅383 main peaks.
It is worth noting that in general the signal reconstruction can be made very economical, and will inherently permit all possible processing gain. For each of the 2N+3 received frequencies:
This method is not restricted to 1-millisecond transmission, and the length of the transmissions may be increased or decreased. However, the total time for all transmissions should be short enough to freeze any motion of the receiver or transmitter.
Obtaining Cancellation on Alternate Half-Cycles of r(t): by simply reversing the polarity of the canceling function g(t), cancellation between the peaks of s(t) is possible where r(t) was formerly oscillatory. However, to obtain cancellation between all peaks of s(t), the function g(t) and its polarity reversed version must be applied at the receiver, and this involves coefficient weighting at the receiver.
Coefficient Weighting at the Receiver: if desired, the coefficients b_{k }in the equation (21) are used for construction of r(t) at the transmitter and may be introduced at the receiver instead. This is easily seen by considering the sequence of signals in the equation (20) in which the final signal is the same if b_{k }is introduced at the last step instead of at the beginning. Ignoring noise, the values are as follows:
cos ω_{k}t (at baseband in transmitter)
cos(ω+ω_{k})t (translated by frequency ω up to RF)
a_{j }cos[(ω+ω_{k})(t−τ_{j})+φ_{j}] (at receiver antenna)
a_{j }cos[ω_{k}(t−τ_{j})+φ_{j}+θ] (translated by frequency−ω to baseband)
a_{j}b_{k }cos[ω_{k}(t−τ_{j})+φ_{j}+θ] (weighted by coefficient b_{k }at baseband) (24)
The transmitter can then transmit all frequencies with the same amplitude, which simplifies its design. It should be noted, that this method also weights the noise at each frequency, the effect of which should be considered. It should also be noted that coefficient weighting should be done at the receiver in order to effect the polarity reversal of g(t) to get twice as many useable main peaks.
Scaling of Δf to Center Frequencies in Channels: to meet the FCC requirements at the VHF or lower frequencies a channelized transmission with constant channel spacing will be required. In a channelized transmission band with constant channel spacing that is small compared to the total allocated band, which is the case for the VHF and lower frequencies band(s), small adjustments to Δf, if necessary, permit all transmitted frequencies to be at channel centers without materially changing performance from original design values. In the two examples of base-band signals previously presented, all frequency components are multiples of Δf/2, so if the channel spacing divides Δf/2, the lowest RF transmitted frequency can be centered in one channel and all other frequencies fall at the center of channels.
In some Radio Frequency (RF)-based identification, tracking and locating systems in addition to performing the distance measurement function, both: the Master Unit and the Tag Unit also perform voice, data and control communication functions. Similarly, in the preferred embodiment both the Master Unit and the Tag perform voice, data and control communication functions in addition to the distance measurement function.
According to the preferred embodiment, the ranging signal(s) are subject to the extensive sophisticated signal processing techniques, including the multi-path mitigation. However, these techniques may not lend themselves to the voice, data and control signals. As a result, the operating range of the proposed system (as well as other existing systems) may be limited not by its ability to measure distance reliably and accurately, but by being out of range during voice and/or data and/or control communications.
In other Radio Frequency (RF)-based identification, tracking and locating systems the distance measurement functionality is separated from the voice, data and control communication functionality. In these systems separate RF Transceivers are used to perform voice, data and control communication functions. The drawback of this approach is system increased cost, complexity, size, etc.
To avoid abovementioned drawbacks, in the preferred embodiment, a narrow bandwidth ranging signal or base-band narrow bandwidth ranging signal several individual frequency components are modulated with the identical data/control signals and in case of voice with digitized voice packets data. At the receiver the individual frequency components that have the highest signal strength are demodulated and the obtained information reliability may be further enhanced by performing “voting” or other signal processing techniques that utilize the information redundancy.
This method allows to avoid the “null” phenomena, wherein the incoming RF signals from multiple paths are destructively combining with the DLOS path and each other, thus significantly reducing the received signal strength and associated with it SNR. Moreover, such method allows to find a set of frequencies at which the incoming signals from multiple paths are constructively combining with DLOS path and each other, thus increasing the received signal strength and associated with it SNR.
As mentioned earlier, spectrum estimation-based super-resolution algorithms generally use the same model: a linear combination of complex exponentials and their complex amplitudes of frequencies. This complex amplitude is given by equation 3 above.
All spectrum estimation-based super-resolution algorithms require a priori knowledge of number of complex exponentials, i.e., the number of multipath paths. This number of complex exponentials is called the model size and is determined by the number of multi-path components L as shown in equations 1-3. However, when estimating path delay, which is the case for RF track-locate applications, this information is not available. This adds another dimension, i.e., the model size estimation, to the spectrum estimation process via super-resolution algorithms.
It has been shown (Kei Sakaguchi et al., Influence of the Model Order Estimation Error in the ESPRIT Based High Resolution Techniques) that in case of model size underestimation the accuracy of frequency estimation is impacted and when the model size is overestimated the algorithm generates spurious, e.g., non-existent, frequencies. Existing methods of model size estimation such as AIC (Akaikes Information Criterion), MDL (Minimum Description Length), etc. have a high sensitivity to correlation between signals (complex exponentials). But in the case of RF multipath, this is always the case. Even, for example, after Forward-Backward smoothing algorithms are applied, there will always be a residual amount of correlation.
In the Sakaguchi paper, it is suggested to use an overestimated model and differentiating actual frequencies (signals) from spurious frequencies (signals) by estimating these signals power (amplitude) and then rejecting the signals with very low power. Although this method is an improvement over existing methods, it is not guaranteed. The inventors implemented the Kei Sakaguchi et al. method and ran simulations for more complex cases with a larger model size. It was observed that, in some cases, a spurious signal may have amplitude that is very close to actual signals amplitude.
All spectrum estimation-based super-resolution algorithms work by splitting the incoming signal complex amplitude data into two sub-spaces: the noise sub-space and signals sub-space. If these sub-spaces are properly defined (separated), then the model size is equal to the signal sub-space size (dimension).
In one embodiment of the present invention, the model size estimation is accomplished using an “F” statistic. For example, for ESPRIT algorithm, the singular value decomposition of the estimate of the covariance matrix (with forward/backward correlation smoothing) is ordered in ascending order. Thereafter, a division is made whereby the (n+1) eigenvalue is divided by the n-th eigenvalue. This ratio is an “F” random variable. The worst case is an “F” random variable of (1,1) degree of freedom. The 95% confidence interval for a “F” random variable with (1,1) degrees of freedom is 161. Setting that value as a threshold determines the model size. Note also that for the noise subspace, the eigenvalues represent an estimate of the noise power.
This method of applying “F” statistics to the ratio of the eigenvalues is a more accurate method of estimating the model size. It should be noted that other degrees of freedom in “F” statistics can be also used for threshold calculation and consequently model size estimation.
Nevertheless, in some cases, two or more very closely spaced (in time) signals can degenerate into one signal because of real-world measurement imperfections. As a result, the above mentioned method will underestimate the number of signals, i.e., the model size. Since model size underestimation reduces the frequency estimation accuracy, it is prudent to increase the model size by adding a certain number. This number can be determined experimentally and/or from simulations. However, when signals are not closely spaced, the model size will be overestimated.
In such cases spurious, i.e., non-existent, frequencies may appear. As noted earlier, using signal amplitude for spurious signals detection does not always work because in some cases a spurious signal(s) was observed to have amplitude that is very close to actual signal(s) amplitude. Therefore, in addition to the amplitude discrimination, filters can be implemented to improve spurious frequencies elimination probability.
The frequencies that are estimated by super-resolution algorithms are artificial frequencies (equation 2). In fact, these frequencies are individual paths delays of the multipath environment. As a result, there should be no negative frequencies and all negative frequencies that are produced by a super-resolution algorithm are spurious frequencies to be rejected.
Furthermore, a DLOS distance range can be estimated from the complex amplitude Â(f_{n}) values obtained during measurements using methods that are different from super-resolution methods. While these methods have lower accuracy, this approach establishes range that is used to discriminate delays, i.e., frequencies. For example, the ratio of
in Δf intervals where the signal amplitude |Â(f_{n})| is close to maximum, i.e., avoiding nulls, provides a DLOS delay range. Although actual DLOS delay can be up to two times larger or smaller, this defines a range that helps to reject spurious results.
In the exemplary embodiment, the ranging signal makes a round-trip. In other words, it travels both ways: from a master/reader to a target/slave and from the target/slave back to the master/reader:
Master transmits a tone: α×e^{−jωt}, where ω is an operating frequency in the operating band and α is the tone signal amplitude.
At the target's receiver, the received signal (one-way) is as follows:
Where: N is number of signal paths in the multipath environment; K0 and τ_{0 }are amplitude and time-of-flight of the DLOS signal; |K_{0}|=1, K_{0}>0, |K_{m≠0}|≦1 and K_{M≠0 }can be positive or negative.
S_{one-way}(t)=α×e^{−jωt}×A(ω)×e^{−jθ(ω)} (26)
Where:
is one way multipath RF channel transfer function in the frequency domain; and A(ω)≧0.
Target retransmits the received signal:
S_{retransmit}(t)=α×e^{−jωt}×A(ω)×e^{−jθ(ω)} (27)
At the master receiver, the round-trip signal is:
Or:
S_{round}_{—}_{trip}(t)=α×e^{−jωt}×A^{2}(ω)×e^{−j2θ(ω)} (28)
On the other hand from equations (26) and (28):
Where:
is roundtrip multipath RF channel transfer function in the frequency domain.
From equation 29, the roundtrip multipath channel has a larger number of paths than one-way channel multipath because the
expression in addition to the τ_{0}÷τ_{N }paths delays, includes combinations of these paths delays, for example: τ_{0}+τ_{1}, τ_{0}+τ_{2 }. . . , τ_{1}+τ_{2}, τ_{1}+τ_{3}, . . . , etc.
These combinations dramatically increase the number of signals (complex exponentials). Hence the probability of very closely spaced (in time) signals will also increase and may lead to significant model size underestimation. Thus, it is desirable to obtain one-way multipath RF channel transfer function.
In preferred embodiment, the one-way amplitude values |Â(f_{n})| are directly obtainable from target/slave device. However, the one-way phase values ∠Â(f_{n}) cannot be measured directly. It is possible to determine the phase of the one-way from the roundtrip phase measurements observation:
However, for each value of ω, there are two values of phase α(ω) such that
e^{j2α(ω)}=e^{jβ(ω)}
A detailed description of resolving this ambiguity is shown below. If the ranging signal different frequency components are close to each other, then for most part of the one-way phase can be found by dividing the roundtrip phase by two. Exceptions will include the areas that are close to the “null”, where the phase can undergo a significant change even with small frequency step. Note: the “null” phenomena is where the incoming RF signals from multiple paths are destructively combining with the DLOS path and each other, thus significantly reducing the received signal strength and associated with it SNR.
Let h(t) be the one-way impulse response of a communications channel. The corresponding transfer function in the frequency domain is
where A(ω)≧0 is the magnitude and α(ω) is the phase of the transfer function. If the one-way impulse response is retransmitted back through the same channel as it is being received, the resulting two-way transfer function is
G(ω)=B(ω)e^{jβ(ω)}=H^{2}(ω)=A^{2}(ω)e^{j2α(ω)} (31)
where B(ω)≧0. Suppose the two-way transfer function G(ω) is known for all ω in some open frequency interval (ω_{1},ω_{2}). Is it possible to determine the one-way transfer function H(ω) defined on (ω_{1},ω_{2}) that produced G(ω)?
Since the magnitude of the two-way transfer function is the square of the one-way magnitude, it is clear that
A(ω)=√{square root over (B(ω))} (32)
However, in trying to recover the phase of the one-way transfer function from observation of G(ω), the situation is more subtle. For each value of ω, there are two values of phase α(ω) such that
e^{j2α(ω)}=e^{jβ(ω)} (33)
A large number of different solutions might be generated by independently choosing one of two possible phase values for each different frequency ω.
The following theorems, which assume that any one-way transfer function is continuous at all frequencies, help resolve this situation.
Theorem 1: Let I be an open interval of frequencies ω containing no zeros of the two-way transfer function G(ω)=B(ω)e^{jβ(ω)}. Let J(ω)=√{square root over (B(ω))}e^{jγ(ω) }be a continuous function on I where β(ω)=2γ(ω). Then J(ω) and −J(ω) are the one-way transfer functions which produce G(ω) on I, and there are no others.
Proof: One of the solutions for the one-way transfer function is the function H(ω)=√{square root over (B(ω))}e^{jα(ω)}, continuous on I since it is differentiable on I, and where β(ω)=2α(ω). Since G(ω)≠0 on I, H(ω) and J(ω) are nonzero on I. Then,
Since H(ω) and J(ω) are continuous and nonzero on I, their ratio is continuous on I, hence the right side of (34) is continuous on I. The conditions β(ω)=2α(ω)=2γ(ω) imply that for each ωεI, α(ω)−γ(ω) is either 0 or π. However, α(ω)=γ(ω) cannot switch between these two values without causing a discontinuity on the right side of (34). Thus, either α(ω)−γ(ω)=0 for all ωεI, or α(ω)−γ(ω)=π for all ωεI. In the first case, we get J(ω)=H(ω), and in the second we get J(ω)=−H(ω).
This theorem proves that to get a one-way solution on any open interval I containing no zeros of the transfer function G(ω)=B(ω)e^{jβ(ω)}, we form the function J(ω)=√{square root over (B(ω))}e^{jγ(ω)}, choosing the values of γ(ω) satisfying β(ω)=2γ(ω) in such a way as to make J(ω) continuous. Since it is known that there is a solution having this property, namely H(ω), it is always possible to do this.
An alternate procedure for finding a one-way solution is based on the following theorem:
Theorem 2: Let H(ω)=A(ω)e^{jα(ω) }be a one-way transfer function and let I be an open interval of frequencies ω containing no zeros of H(ω). Then the phase function α(ω) of H(ω) must be continuous on I.
Proof: Let ω_{0 }be a frequency in the interval I. In FIG. 7, the complex value H(ω_{0}) has been plotted as a point in the complex plane, and by hypothesis, H(ω_{0})≠0. Let ε>0 be an arbitrarily small real number, and consider the two angles of measure ε shown in the FIG. 7, as well as the circle centered at H(ω_{0}) and tangent to the two rays OA and OB. By assumption, H(ω) is continuous for all ω. Thus, if ω is sufficiently close to ω_{0}, the complex value H(ω) will lie in the circle, and it is seen that |α(ω)−α(ω_{0})|<ε. Since ε>0 was chosen arbitrarily, we conclude that α(ω)→α(ω_{0}) as ω→ω_{0}, so that the phase function α(ω) is continuous at ω_{0}.
Theorem 3: Let I be an open interval of frequencies ω containing no zeros of the two-way transfer function G(ω)=B(ω)e^{jβ(ω)}. Let J(ω)=√{square root over (B(ω))}e^{jγ(ω) }be a function on I where β(ω)=2γ(ω) and γ(ω) is continuous on I. Then J(ω) and −J(ω) are the one-way transfer functions which produce G(ω) on I, and there are no others.
Proof: The proof is similar to the proof of Theorem 1. We know that one of the solutions for the one-way transfer function is the function H(ω)=√{square root over (B(ω))}e^{jα(ω)}, where β(ω)=2α(ω). Since G(ω)≠0 on I, H(ω) and J(ω) are nonzero on I. Then,
By hypothesis γ(ω) is continuous on I and by Theorem 2 α(ω) is also continuous on I. Thus, α(ω)−γ(ω) is continuous on I. The conditions β(ω)=2α(ω)=2γ(ω) imply that for each ωεI, α(ω)−γ(ω) is either 0 or π. However, α(ω)−γ(ω) cannot switch between these two values without becoming discontinuous on I. Thus, either α(ω)−γ(ω)=0 for all ωεI, or α(ω)−γ(ω)=π for all ωεI. In the first case, we get J(ω)=H(ω), and in the second J(ω)=−H(ω).
Theorem 3 tells us that to get a one-way solution on any open interval I containing no zeros of the transfer function G(ω)=B(ω)e^{jβ(ω)}, we simply form the function J(ω)=√{square root over (B(ω))}e^{jγ(ω)}, choosing the values of γ(ω) satisfying β(ω)=2γ(ω) in such a way as to make the phase function γ(ω) continuous. Since it is known that there is a solution having this property, namely H(ω), it is always possible to do this.
Although the above theorems show how to reconstruct the two one-way transfer functions which generate the two-way function G(ω), they are useful only on a frequency interval I containing no zeros of G(ω). In general, G(ω) will be observed on a frequency interval (ω_{1}, ω_{2}) which may contain zeros. The following is a method that might get around this problem, assuming that there are only a finite number of zeros of G(ω) in (ω_{1}, ω_{2}), and that a one-way transfer function has derivatives of all orders on (ω_{1}, ω_{2}), not all of which are zero at any given frequency ω:
Let H(ω) be a one-way function that generates G(ω) on the interval (ω_{1}, ω_{2}), and assume that G(ω) has at least one zero on (ω_{1}, ω_{2}). The zeros of G(ω) will separate (ω_{1}, ω_{2}) into a finite number of abutting open frequency intervals J_{1}, J_{2}, . . . , J_{n}. On each such interval the solution H(ω) or −H(ω) will be found using either Theorem 1 or Theorem 3. We need to “stitch together” these solutions so that the stitched solution is either H(ω) or −H(ω) across all of (ω_{1}, ω_{2}). In order to do this, we need to know how to pair the solutions in two adjacent subintervals so that we aren't switching from H(ω) to −H(ω) or from −H(ω) to H(ω) in moving from one subinterval to the next.
We illustrate the stitching procedure starting with the first two adjacent open subintervals J_{1 }and J_{2}. These subintervals will abut at a frequency ω_{1 }which is a zero of G(ω) (of course, ω_{1 }is not contained in either subinterval). By our above assumption about the properties of a one-way transfer function, there must be a minimum positive integer n such that H^{(n)}(ω_{1})≠0, where the superscript (n) denotes the n^{th }derivative. Then the limit of the n^{th }derivative of our one-way solution in J_{1 }as ω→ω_{1 }from the left will be either H^{(n)}(ω_{1}) or −H^{(n)}(ω_{1}) according to whether our solution in J_{1 }is H(ω) or −H(ω). Similarly, the limit of the n^{th }derivative of our one-way solution in J_{2 }as ω→ω_{1 }from the right will be either H^{(n)}(ω_{1}) or −H^{(n)}(ω_{1}) according to whether our solution in J_{2 }is H(ω) or −H(ω). Since H^{(n)}(ω_{1})≠0, the two limits will be equal if and only if the solutions in J_{1 }and J_{2 }are both H(ω) or both −H(ω). If the left and right hand limits are unequal, we invert the solution in subinterval J_{2}. Otherwise, we don't.
After inverting the solution in subinterval J_{2 }(if necessary), we perform an identical procedure for subintervals J_{2 }and J_{3}, inverting the solution in subinterval J_{3 }(if necessary). Continuing in this fashion, we eventually build up a complete solution on the interval (ω_{1}, ω_{2}).
It would be desirable that high-order derivatives of H(ω) not be required in the above reconstruction procedure, since they are difficult to compute accurately in the presence of noise. This problem is unlikely to occur, because at any zero of G(ω) it seems very likely that the first derivative of H(ω) will be nonzero, and if not, very likely that the second derivative will be nonzero.
In a practical scheme, the two-way transfer function G(ω) will be measured at discrete frequencies, which must be close enough together to enable reasonably accurate computation of derivatives near the zeros of G(ω).
For RF-based distance measurements it is necessary to resolve an unknown number of closely spaced, overlapping, and noisy echoes of a ranging signal with a priori known shape. Assuming that ranging signal is a narrow-band, in frequency domain this RF phenomena can be described (modeled) as a sum of a number of sine waves, each per multipath component, and each with the complex attenuation and propagation delay of the path.
Taking the Fourier transform of the above mentioned sum will express this multipath model in the time domain. Exchanging the role of time and frequency variables in this time domain expression, this multipath model will become harmonic signals spectrum in which the propagation delay of the path is transformed to a harmonic signal.
The super (high) resolution spectral estimation methods are designed to distinguish closely-placed frequencies in the spectrum and used for estimating the individual frequencies of multiple harmonic signals, e.g., paths delays. As a result, path delays can be accurately estimated.
The super resolution spectral estimation makes use of the eigen-structure of the covariance matrix of the baseband ranging signal samples and covariance matrix intrinsic properties to provide a solution to an underlying estimation of individual frequencies, e.g. paths delays. One of the eigen-structure properties is that the eigenvalues can be combined and consequently divided into orthogonal noise and signal eigenvectors, aka subspaces. Another eigen-structure property is the rotation-invariant signal subspaces property.
The subspace decomposition technology (MUSIC, rootMUSIC, ESPRIT, etc.) relies on breaking the estimated covariance matrix of the observed data into two orthogonal subspaces, the noise subspace and the signal subspace. The theory behind the subspace decomposition methodology is that the projection of the observable onto the noise subspace consists of only the noise and the projection of the observable onto the signal subspace consists of only the signal.
The spectral estimation methods assume that signals are narrow-band, and the number of harmonic signals is also known, i.e., the size of the signal subspace needs to be known. The size of the signal subspace is called as the model size. In general, it cannot be known in any detail and can change rapidly—particularly indoors—as the environment changes. One of the most difficult and subtle issues when applying any subspace decomposition algorithm is the dimension of the signal subspace that can be taken as the number of frequency components present, and which is the number multipath reflections plus the direct path. Because of real-world measurement imperfections there always will be an error in the model size estimation, which in turn will result in loss of accuracy of frequencies estimation, i.e., distances.
To improve the distance measurement accuracy, one embodiment of the present invention includes six features that advance the state of the art in the methodology of subspace decomposition high resolution estimation. Included is combining two or more algorithms estimating individual frequencies by using different eigen-structure properties that further reduces the delay path determination ambiguity.
Root Music finds the individual frequencies, that when the observable is projected onto the noise subspace, minimizes the energy of the projection. The Esprit algorithm determines the individual frequencies from the rotation operator. And in many respects this operation is the conjugate of Music in that it finds the frequencies that, when the observable is projected onto the signal subspace, maximizes the energy of the projection.
The model size is the key to both of these algorithms, and in practice, in a complex signal environment such as seen in indoor ranging—the model size which provides the best performance for Music and Esprit are in general not equal, for reasons that will be discussed below.
For Music it is preferable to err on the side of identifying a basis element of the decomposition as a “signal eigen value” (Type I Error). This will minimize the amount of signal energy that is projected on the noise subspace and improve the accuracy. For Esprit—the opposite is true—it is preferable to err on the side of identifying a basis element of the decomposition as a “noise eigenvalue.” This is again a Type I Error. This will minimize the impact of noise on the energy projected onto the signal subspace. Therefore, the model size for Music will, in general, be somewhat larger than that for Esprit.
Secondly, in a complex signal environment, there arise occasions where, with the strong reflections and the potential that the direct path is in fact much weaker than some of the multipath reflections, the model size is difficult to estimate with sufficient statistical reliability. This issue is addressed by estimating a “base” model size for both Music and Esprit and the processing the observable data using Music and Esprit in a window of model sizes defined by the base model size for each. This results in multiple measurements for each measurement.
The first feature of the embodiment is the use of the F-statistic to estimate the model size (see above). The second feature is the use of different Type I Error probabilities in the F-statistic for Music and Esprit. This implements the Type I Error differences between Music and Esprit as discussed above. The third feature is the use of a base model size and a window in order to maximize the probability of detecting the direct path.
Because of the potentially rapidly changing physical and electronic environment, not every measurement will provide robust answers. This is addressed by using cluster analysis on multiple measurements to provide a robust range estimate. The fourth feature of the embodiment is the use of multiple measurements.
Because there are multiple signals present, the probability distribution of the multiple answers resulting from multiple measurements, each using multiple model sizes from both a Music and Esprit implementation, will be multimodal. Conventional cluster analysis will not be sufficient for this application. The fifth feature is the development of multimodal cluster analysis to estimate the direct range and equivalent range of the reflected multipath components. The sixth feature is the analysis of the statistics of the range estimates provided by the cluster analysis (range and standard deviation and combing those estimates that are statistically identical. This results in a more accurate range estimate.
The abovementioned method can be also used in wide bandwidth ranging signal location-finding systems.
Having thus described the different embodiments of a system and methods, it should be apparent to those skilled in the art that certain advantages of the described method and apparatus have been achieved. In particular, it should be appreciated by those skilled in the art that a system for tracking and locating objects can be assembled using FGPA or ASIC and standard signal processing software/hardware combination at a very small incremental cost. Such a system is useful in a variety of applications, e.g. locating people in indoor or in outdoor environments, harsh and hostile environments etc.
It should also be appreciated that various modifications, adaptations, and alternative embodiments thereof may be made within the scope and spirit of the present invention. The invention is further defined by the following claims.
Appendix: Derivation of r(t) in the thresholded method:
Starting with expression (20), we obtain
where the trigonometric identity sin x sin y=½ cos(x−y)−½ cos(x+y) is used.
Except for α_{a}, the coefficients α_{k }are zero for even k. The reason for this is that on the interval I, the function 1/sin πΔft that we are tying to approximate by h(t) is even about the center of I, but the basis functions sin kπΔft for even k, k≠0, are odd about the center of I, hence are orthogonal to 1/sin πΔft on I. Thus, we can make the substitution k=2n+1 and let M be an odd positive integer. In fact, we will let M=2N+1. This choice has been experimentally determined to provide a good amount of cancellation of the oscillations in the interval I.
Now we make the substitution k=N−n in the first summation and k=N+n+1 in the second summation to obtain
Subtracting g(t) from s(t) results in
Now let
Then (A4) can be written as