Jam-resistant radio system

Abstract

A radio architecture for communication between transceivers. The first transceiver has a frequency-tunable, free running oscillator connected between a signal port and a first signal processing module. Aa first control module is connected to the free running oscillator that tunes the free running oscillator such that, in a transmit mode, the free running oscillator generates a transmit signal that includes a message signal and a hopping signal. The transmit signal is continuously smooth and the hopping signal is pseudorandom. In a receive mode, the free running oscillator generates the hopping signal for use by the first signal processing module to extract a received message signal from a receive signal. The second transceiver has a second signal processing module and a tunable oscillator. A second control module has instructions to frequency-match the tunable oscillator to the free running oscillator using the hopping signal.

Description

JAM-RESISTANT RADIO SYSTEMTECHNICAL FIELD

[0001] This relates to a radio architecture that uses a transceiver, and in particular, a frequency-hopping transceiver.BACKGROUND

[0002] In a world where remote sensors and remote vehicles are increasingly deployed in contended spaces, the issue of jamming of such sensors and remote vehicles by the opposing force (OPFOR) presents a threat for both control of, and information transfer, from such sensors and remote vehicles which are assumed mobile as a worst case. In contended spaces, OPFOR jammers are generally located around an asset that the opposing force desires to protect. These jammers are relatively short range, in the ranges of kilometers, and are relatively low power compared to larger, more powerful area jammers. Given that short range jammers are lower power, they are less expensive and more readily deployed.

[0003] The mitigation of such jammers has become a critical objective in contended spaces. This disclosure teaches a method to effectively mitigate such jamming, whether local, low power jammers, or more powerful area jammers that are farther from the local asset to be protected by the OPFOR. As such, the sensor / remote vehicle is at a distance from the OPFOR target of interest.

[0004] An overview of a typical asset protecting jammer scenario is presented in FIG. 1, where the OPFOR jammer 102 is located within 100 meters of the protected asset 100, and transmitting at 100 W. The remote vehicle (RV) control transceiver 104 is located at the ground control station (GCS) at some 5 km from the remote vehicle 100. Any remote moving platform may be considered an RV, whether a ground vehicle, airborne platform, or a moving person.

[0005] The OPFOR jammer 100 is attempting to identify the frequency of the control and information signals to jam both control and data. When the control and data signals are continuously frequency hopping at a high rate and over a broad frequency range, there is processing gain at both CGS and RV 102 that overcomes any jamming activity.

[0006] The GCS control is manned by trained operators who are supplied with first person video (FPV) from the RV. This direct view of the battlespace enables operators to effectively control the RV in real time. Disrupting this FPV for the CGS operators is a key objective of OPFOR jamming.

[0007] The FPV control of a RV requires both the pilot to RV (P2D) down link with command signals and the video uplink RV to pilot (D2P) to be operational. The rate of the signal degree of freedom (DOF) for the command link only needs to be a very low rate of about 1kb / sec. (yaw, pitch, roll commands at about 0.1 sec intervals + control switches). The D2P video link can be < 1Mbps and still have sufficient information content such that a trained pilot can do the FPV. An assumption is that the jammer will be located close to the target, say, about 1 km from the command post with the pilot. Hence the RV in the last few tens of meters from the target will experience a large jamming content in the P2D link. For the D2P link, the jammer will compete directly with the RV video transmission back to the command post.

[0008] In many examples, it may be assumed that jammer 100 will not attempt to follow the continuous frequency hopping of the D2P and P2D emissions but will have uniform spectral power density over the hopping bandwidth which could be upwards of 1 GHz. The ideal processing gain of the P2D and D2P links is the degree of freedom (DOF) per second that the jammer must transmit based on the hopping bandwidth to the DOF / sec of the P2D link (about 1kb / sec) and the D2P link (about 1Mbps). Mapping from kb / sec to DOF / sec requires some analysis of the coding and information theory which will not be discussed herein. The video link may also be augment with RV-based CV (computer vision), Al, etc. As these relate to managing the data signal, they will not be considered further herein.

[0009] An object of a jam resistant link for a simple jammer may be to maximise the DOF ratio, which may be achieved by increasing the hopping bandwidth and by increasing the speed of the hopping.

[0010] In both links, spreading the P2D and D2P data over the 1 GHz bandwidth will provide the jammer suppression needed. Suppose a simple BPSK modulation is used then the DOF needed is approximately the bit rate. The Shannon limit for Eb / No is -1.6 dB so it is notunreasonable to simply approximate the Eb / No requirement for the coded BPSK link as 0 dB. Then the EIRP required of the jammer to compete with the RV is then simply the 1 GHz bandwidth divided by the DOF / sec of the link which is approximately equivalent to the bit rate.

[0011] Hence for the D2P link, there may be a processing suppression of the jammer signal of 30 dB (assuming 1 GHz hopping and 1 Mbps video). For the P2D link, there may be a jammer suppression of 60 dB (assuming a 1 GHz hopping and a 1 kbps command link). This suppression is key to providing adequate robustness of the data comms in the context of the jammer.

[0012] An obvious implementation that will achieve this jammer suppression is to have a high speed spread spectrum modulation of the data. Suppose for the command link the BPSK bit period is T=1msec. Then the spread spectrum chipping rate can be 1 GHz. Then there will be 1 e6 chips generated for each bit. After the bit epoch the chipping sequence can change in a pseudo random fashion and the next bit propagated. The chipping symbols can be formed such that they have the correct spectrum of say covering the band between 2 to 3 GHz. Also there would be a different chipping sequence for a ‘0’ or T bit.

[0013] The overall potential arbitrary waveform generator (AWG) link processing is shown in FIG. 2. Here it is assumed an AWG 200 that simply consists of a high-speed DAC that is fed from a digital word sequence in memory that represents the chipping samples for the 0 or 1 bit of the data. The receiver 202 demodulates based on a correlator 204 with a matching AWG 200 for ‘0’ and for T denoted as AWG0 and AWG1 respectively.

[0014] The correlator outputs are passed to detectors 206 which in the ideal case just measure the amplitude of the correlator output.

[0015] Additionally, time synchronization is required as the RV is moving and the clock in the RV is not locked to the timing clock in the command post. Synchronization can be done with conventional means, such as an early late gate as described elsewhere.

[0016] So, in principle it is possible to build a transceiver pair that can achieve the jammer suppression desired. The chipping signals may be designed such that they are impossible forthe jammer to guess or track or quickly adapt to. Hence the jammer is forced to tailor its emission to fit the averaged power spectral density (PSD) of the P2D and D2P links. For the P2D this can realistically provide 60 dB of jammer suppression and for the D2P about 30 dB.

[0017] What is less practical is the implementation of the AWG for the RV transceiver. Despite advances in digital chip technology, sample generation at a 2 GHz rate is expensive and consumes high levels of power, making it costly to build into an expendable RV.SUMMARY

[0018] According to an aspect, there is provided a radio architecture, comprising a first transceiver in wireless communication with a second transceiver. The first transceiver comprises a first antenna, a first signal processing module connected to the antenna, a free running oscillator connected to the antenna and the first signal processing module and that is frequency tunable, and a first control module connected to the free running oscillator. The first control module comprises instructions to tune the free running oscillator such that, in a transmit mode, the free running oscillator generates a transmit signal that comprises a message signal and a hopping signal, wherein the transmit signal is continuously smooth and the hopping signal is pseudorandom. In a receive mode, the free running oscillator generates the hopping signal for use by the first signal processing module to extract a received message signal from a receive signal. The second transceiver comprises a second antenna, a second signal processing module connected to the antenna, a tunable oscillator connected to the second signal processing module, and a second control module that comprises instructions to frequencymatch the tunable oscillator to the free running oscillator using the hopping signal.

[0019] According to other aspects, the free running oscillator may comprise a voltage controlled oscillator or a Q-enhanced resonator; the first transceiver may be an airborne transceiver and the second transceiver may be a ground-based transceiver; the first transceiver may be a mobile transceiver and the second transceiver may be a base transceiver; the second transceiver may comprise a second free running oscillator; the first control module and the second control module may be programmed with instructions to cause the first transceiver andthe second transceiver to communicate using time division duplex; the hopping signal may comprise a central frequency that changes within a hopping bandwidth; the first signal processing module may comprise a mixer and, in the receive mode, the free running oscillator is adapted to act as a local oscillator for the mixer; in the second control module, the instructions to frequency-match the tunable oscillator may further comprise calculating a warping function that represents a difference between an expected frequency and an actual frequency of the free running oscillator; the warping function may be adapted to correct for at least a frequency drift of the free running oscillator and a transmission time delay between the first transceiver and the second transceiver; the warping function may be adjusted based on a calibration signal transmitted between the transmit signal and the receive signal, the calibration signal comprising the hopping signal in isolation; the first signal processing module may comprise at least one signal filter and a demodulator; and / or the first signal processing module may comprise a phased locked loop (PLL).

[0020] According to an aspect, there is provided a method of wireless communication, comprising the steps of:providing a first transceiver that comprises a first antenna, a first signal processing module connected to the antenna, a free running oscillator connected to the antenna and the first signal processing module where the free running oscillator being frequency tunable, and a first control module connected to tune the free running oscillator;providing a second transceiver that comprises a second antenna, and a signal processing module connected to the antenna, the signal processing module comprising a tunable oscillator;in the first transceiver, tuning the free running oscillator such that, in a transmit mode, the free running oscillator generates a transmit signal that comprises a message signal and a hopping signal, the hopping signal comprising a continuously smooth pseudorandom signal, and in a receive mode, the free running oscillator downconverts a receive signal received via the antenna using the hopping signal; andin the second transceiver, tuning the tunable oscillator to frequency-match the tunable oscillator to the free running oscillator.

[0021] According to other aspects, the method may comprise one or more of the following features: the free running oscillator may comprise a voltage controlled oscillator or a Q-enhanced resonator; the first transceiver may be an airborne transceiver, and the second transceiver is a ground-based transceiver; the first transceiver may be a mobile transceiver and the second transceiver is a base transceiver; the second transceiver may comprise a second free running oscillator; the method may further comprise the step of using time division duplex to communicate between the first transceiver and the second transceiver; the tunable oscillator may be tuned based on the hopping signal, the hopping signal comprising a central frequency that changes within a hopping bandwidth; the first transceiver comprises a mixer and, in the receive mode, the free running oscillator acts as a local oscillator; in the second control module, frequency-matching the tunable oscillator may comprise the use of a warping function that represents a difference between an intended frequency and an actual frequency of the free running oscillator; the method may further comprise the step of updating the warping function based on a received signal from the first transceiver; the warping function may be stored in a look up table; the warping function may be adapted to correct for frequency drift of the free running oscillator and transmission time delay; the warping function may be adjusted based on a calibration signal transmitted between the transmit signal and the receive signal; in the transmit mode, the first control module may maintain a coherent phase as the free running oscillator is tuned between frequencies; the first signal processing module may comprise at least one signal filter and a demodulator; and / or the demodulator may further comprise a phased locked loop (PLL).

[0022] According to an aspect, there is provided a radio architecture, comprising a first transceiver in communication with a second transceiver; wherein the first transceiver comprises a first signal processing module connected to receive a receive signal, a free running oscillator connected to the first signal processing module where the free running oscillator being frequency tunable, a first control module connected to the free running oscillator. The first control module comprises instructions to tune the free running oscillator such that, in a transmit mode, the free running oscillator generates a transmit signal to the second transceiver that comprises a message signal and a hopping signal, wherein the transmit signal is continuously smooth and the hopping signal is pseudorandom; and in a receive mode, the free running oscillator generates the hopping signal for use by the first signal processing module to extract a received message signal from the receive signal. The second transceiver comprises a second signal processing module connected to receive the transmit signal from the first transceiver, a tunable oscillator connected to the second signal processing module, and a second control module thatcomprises instructions to frequency-match the tunable oscillator to the free running oscillator using the hopping signal.

[0023] According to an aspect, there is provided a radio architecture in which a first transceiver comprises a resonator that alternates between operating in self-oscillation mode (SOM) and as a Q-enhanced resonator (QER), and a second transceiver characterizes the output. This may be used to implement a time division duplexing (TDD) communication system. The resonator may operate in SOM and is characterized by the second transceiver but stays in SOM and may be used as a LO with an IF offset to avoid quadrature demodulation. The IF offset may be implemented by the first transceiver, or may be accommodated in the signal transmitted by the second transceiver.

[0024] According to an aspect, there is provided a first transceiver system comprising a free running voltage-controlled oscillator (VCO) that is frequency modulated by a control module for transmit mode and used as a down-conversion LO in receive mode with the output of the down conversion into an FM demodulator, which is paired to a second transceiver system of equivalent architecture, wherein the VCO of the second transceiver system is continuously adjusted to match the frequency tuning of the first system. A control module in the first system to control the TDD mode and message modulation. A control module in the second system to facilitate the run time adaptation of the matching of the tuning characteristics of the VCO to the VCO of the first system.

[0025] According to other aspects, the the first system, in transmit mode, may have a VCO modulation of h+m where h is a pseudorandom, arbitrary hopping signal that is mutually known with system 1 and system 2 and a message signal m. In the second system, in receive mode, the VCO modulation is w(h) where w is a warping function that is continuously adjusted by sensing the modulated transmission from the first system; the first system in receive mode may have a VCO modulation of h where h is an arbitrary hopping signal that is mutually known with system 1 and system 2. the VCO of the second system may be modulated by w(h+m) where w() is the warping function; the FM control module may be a first IF followed by a PLL used as an FM demodulator, or may be a first IF followed by a second IF followed by a PLL.

[0026] In other aspects, the features described above may be combined together in anyreasonable combination as will be recognized by those skilled in the art.BRIEF DESCRIPTION OF THE DRAWINGS

[0027] These and other features will become more apparent from the following description in which reference is made to the appended drawings, the drawings are for the purpose of illustration only and are not intended to be in any way limiting, wherein:FIG. 1 is a schematic diagram of a typical asset jammer scenario.FIG. 2 is a schematic diagram of an arbitrary waveform generator link.FIG. 3a is a schematic diagram of the operation of a remote vehicle (RV) to ground control station (GCS).FIG. 3b is a schematic diagram of the operation of a GCS to RV downlink. FIG. 4 is a schematic diagram of a free running dual mode VCO transceiver (FRVT). FIG. 5 is a schematic diagram of a remote vehicle transmitting to a ground control station.FIG. 6 is a schematic diagram of a remote vehicle receiving from a ground control station.FIG. 7 is a schematic diagram of a ground control station.FIG. 8 is a schematic diagram of a resonator in QER mode.FIG. 9 is a schematic diagram of a resonator in Self Oscillation Mode (SOM).FIG. 10 is a schematic diagram of a VCO in switched mode in the Q-enhanced resonator (QER) loop.FIG. 11 is a plot of oscillator frequencies as a function of control voltage variation. FIG. 12 is a schematic diagram of a transceiver link between an FM transmitter and an FM receiver.FIG. 13 is a schematic diagram of a transceiver link between an FM transmitter and the FM receiver.FIG. 14 is a block diagram of a PLL with an input frequency u1(h+m).FIG. 15 is a block diagram of a PLL with a calibrated offset.FIG. 16 is a transceiver connected to a PLC data cable and antenna RF link for prelaunch configuration and calibration.FIG. 17 is a block diagram of an RV Receiver.FIG. 18 is a block diagram of a quadrature ZIF.FIG. 19 is a block diagram of an RV transceiver.FIG. 20 is a block diagram of a GCS transceiver.FIG. 21 is a graph of a detector response where the input frequence of is too high. FIG. 22 is a graph of a detector response where the input frequency is correct. FIG. 23 is a graph of a detector response where the input frequency is too low. FIG. 24 is a block diagram of a GCS upling with time alignment.FIG. 25 is a block diagram illustrating propagation delay in a downlink.FIG. 26 is a block diagram of a GCS PLL.FIG. 27 is a graph of phase over time related to m(t) and h(t).FIG. 28 is a block diagram of a receiver configured for frequency modulation.FIG. 29 is a block diagram of an transceiver using an oscillator as a mixer.FIG. 30 is a VCO calibration pulse diagram.FIG. 31 is a bloc diagram of a PLL configured to be calibrated.FIG. 32 is an example of a look-up-table (LUT) containing a set of samples and interpolations.FIG. 33 is a block diagram depicting the tracking of tdBusing a PLL.FIG. 34 is a block diagram of a PLL aided with an estimate of h delayed based on g(t’)-tdB.FIG. 35 is a response of a discriminator.FIG. 36 is a model of transition matrices model.FIG. 37 is a block diagram of an uplink transmission.FIG. 38 is a block diagram of a downlink transmission.FIG. 39 is a diagram of a TDD structure.FIG. 40 is a block diagram of a transmission link with message FM modulation. FIG. 41 is a graph showing the results of developing a fitted data model for noisy data.FIG. 42 is a graph depicting the frequency deviation from LS curve fitting.FIG. 43 is a graph depicting the calculated TCXO Allen Deviation as a function of various τ values.FIG. 44 is a graph depicting the results of a Kalman Filter for position estimation. FIG. 45 is a graph depicting the translation of a tuning curve due to changes in temperature and supply voltage.FIG. 46 is a graph that depicts the process of the setting the center trackingfrequency.FIG. 47 is a block diagram of an RV in transmit mode.FIG. 48 is a block diagram of an RV in receive mode.DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS

[0028] There will now be described a radio architecture 10 that allows communication between a first transceiver 12 and second transceiver 14. The transceivers 12 and 14 may be in wireless communication, in which case each transceiver includes an antenna 16. Radio architecture 10 may also be used for wired applications, although certain benefits of the architecture described herein may not be realized in a wired application. Where used for a wired application, the antenna may be replaced with a signal connection port. With this in mind, the discussion below will be in the context of a wireless communication architecture.

[0029] Each transceiver includes a signal processing module and a control module. The signal processing module and control module may be separate components, or may be incorporated into the same component, such as a suitably designed processor(s) or circuit board(s).

[0030] The first transceiver includes a free running oscillator that is connected to the antenna and the first signal processing module. The free running oscillator is frequency tunable. The control module of the first transceiver a first control module connected to the free running oscillator. The control module is programmed to tune the free running oscillator such that, in a transmit mode, the free running oscillator generates a transmit signal that comprises a message signal and a hopping signal, and in a receive mode, the free running oscillator generates the hopping signal for use by the first signal processing module to extract a received message signal from a receive signal. The transmit signal generated by the control module and the free running oscillator is continuously smooth, while the hopping signal is pseudorandom. The message and hopping signal will be described in more detail below.

[0031] The second transceiver will also have a signal processing block and a tunable oscillator, and a control module. The second transceiver may be the same as the first transceiver or may be different as described below. In particular, the first transceiver may be used as partof a mobile platform, such as an airborne platform, that may be designed with lower cost, lower power requirements, etc. where a ground-based or fixed platform may not face the same design limitations. As will be described in more detail below, the free running oscillator may be a voltage controlled oscillator or a Q-enhanced resonator.

[0032] As will be described below, radio architecture 10 may be used to provide a high level of jam resistance while still maintaining full frequency division duplex (FDD) operation with insertion an RV 100 at low cost. The ground control station (GCS) 104 is outfitted with an interface module enabling upgrade to the radio architecture 10 at low initial cost and no recurring cost.

[0033] The discussion will be given in the context of a jam-resistant radio as discussed in the Background section, such as may be used between a mobile remote vehicle 100 and a stationary base station. However, it will be understood that radio architecture 10 described herein may be applied to other use scenarios, and that any discussion of a specific use or design objective is provided by way of an example only and is not intended to limit the scope of a specific design element. Furthermore, the intentions and objectives described herein are merely considerations that may be applicable in some, but not all, use scenarios.

[0034]

[0035] Radio architecture 10 described herein involves radios that employ resonator elements in both RV 100 and CGS 104. Regular or continuous calibration may be used to improve the quality of communication between RV 100 and CGS 104. Coherent FM modulation may be used in the radio link to improve the time maintenance between RV 100 and the GCS 104.

[0036] Radio architecture 10 provides a two-way communication system between GCS 104 and RV 100 that may improve hardening against possible jamming. Architecture 10 may be employed using asymmetrical time division duplexing (TDD) where the average uplink is in the range of 1 Mbps and the downlink is several kbps.

[0037] Jammer hardening may be achieved using frequent or continuous frequency hoppingthat is sufficiently fast and over a broad frequency range to force the jammer 102 to transmit over a large frequency range. The effectiveness of the jamming suppression may be quantified as 10log(frequency-hopping-range / rate-of-DOF-of-comm-link). The hopping may be done with a continuous wave (CW) analog frequency modulated (FM) signal. In this example, the hopping waveform may be characterized by a continuous phase as opposed to discrete steps.

[0038] In many circumstances, it is beneficial to implement radio architecture 10 with a low cost, low power, small size RV transceiver. In some examples, the radio architecture may involve a pseudo random hopping over a sufficiently large frequency range, such as 3-4 GHz, at a sufficiently short hop interval, such as on the order of 1 μsec. In one implementation, the frequency hopping may be enabled using a free running oscillator that is tunable. In the discussion below, the free running oscillator will be discussed in terms of a VCO with analog frequency control, however other tunable, free running oscillators may be used. The tuning characteristic of a free running VCO may vary based on external characteristics, such as temperature, aging and / or small variations in the supply voltage. These variances may be tracked and accommodated for by the GCS, which may use remote characterization / modelling of the RV VCO. The modelling may be initiated during initial Pre-Launch Calibration & Configuration (PLC) and / or maintained during run time. With the analog VCO, the FM modulation depth and rate may be based on the varactor bias circuitry, which can easily have a bandwidth of several MHz.

[0039] The GCS may be modelled based on the TDD operation. The use of TDD allows for asymmetry between the uplink and the downlink. In one example, the TDD may alternate between a longer uplink and shorter downlink epoch in a cycle that repeats at a rate of about 30 TDD cycles per second. In this example, the FPV video will be at 30 frames per second and the command updated every 33 msec, which may be adequate for some situations, such as where the RV has sluggish flight dynamics, which may be the cases where the RV is a flying drone carrying a relatively heavy ordinance. In this examples, the TDD scheme allows for a modelling update of the VCO every 33 msec, which may sufficient if the variation of the VCO tuning characteristic is slower than this. In some situations, it may be useful to use the same VCO in the RV and GCS, and initiate the model on the assumption that the VCOs are identical.

[0040] FIG. 3a and FIG. 3b show a block diagram of the basic operation of an example ofan RV and GCS for uplink and downlink, respectively, where m(t) is the data message signal that is encoded into an analog waveform and h0(t) is the pseudo-random continuous FM frequency hopping waveform. Both the uplink and downlink have a propagation delay Dp.

[0041] In this example, for the uplink shown in FIG. 3a, the RV VCO 20 of RV 100 is modulated with m(t)+h0(t) and receiver VCO 20 of GCS 104 with h0(t). In this way a receiver 22 can be implemented which extracts the message signal m(t). For the downlink shown in FIG.3b, the order is reversed such that VCO 20 of GCS 104 is modulated with m(t)+h0(t) and receiver VCO 20 of RV 100 with h0(t).

[0042] The VCOs 20 used by RV and GCS may differ, and the time t may not be exactly the same in the RV and GCS. First, consider the time generated in the RV by a temperature compensated crystal oscillator (TCXO) with a counter circuit generating a local t at the RV. There is structure in the VCO modulated signal that is received by the GCS such that t can be extracted. The GCS has its own TCXO and clock counter generating a local time of t’. Then the time difference of (t-t’) is determined and used such that the hopping and message signal of h0(t)+m(t) may be time synchronized.

[0043] The VCO of the GCS may be aligned with the RV VCO by modelling the difference in the frequency tuning curves, which may be captured by an interpolation function of g(). This allows the GCS VCO to be modulated with a function of g(h0(t)) for uplink receiver operation and g(ho(t)+m(t)) for downlink data transmission. In this way the RV VCO may be free running and the GCS VCO is adjusted to match the RV VCO. This may be accomplished by maintaining a model of the difference of the RV and GCS VCO frequency modulation characteristic captured by g(), such as a warping model. The input for maintaining g() is available from the structure contained in the RV signal that is received by the GCS receiver.

[0044] In the discussion below, various models and considerations for practical applications are described. It will be understood that these are given to provide a more thorough understanding of the architecture and are non-limiting. In particular, it will be understood that these considerations may vary where different components, different operating conditions, or different specifications are used.

[0045] Free Running Dual Mode VCO Transceiver (FRVT)

[0046] An example of a radio architecture will now be discussed in more detail, in which the core resonator of the RV transceiver is a stable Q-enhanced resonator (QER) that relies upon active positive feedback to orthogonally control both center frequency and bandwidth. This orthogonal control provides the requisite stability of the active feedback resonator. An example of such a QER resonator is found in United States patent no. 10,050,604 (Nielsen) entitled “Variable Filter.

[0047] In this example, the resonator may be chosen to be low-cost, low power, small size, and light weight that is suitable to be incorporated into an expendable RV of limited life, and may be a free running Voltage Controlled Oscillator (VCO). The term "free-running" indicates that it operates independently, producing a steady oscillation at a frequency determined by its design and the applied control voltage, without the need for an external trigger signal to start and / or continue oscillating. The VCO can be FM modulated with fast hopping and a message signal.

[0048] Consider a resonator with positive feedback for Q enhancement. It may be a bandpass filter operated as an oscillator or a VCO.

[0049] The RV transceiver may involve a relatively simple circuit that may have fast phase coherent FM modulation in a transmitter mode. In a receiver mode, the VCO may be used as a down conversion LO for a received signal with a fast hopping carrier signal modulated by a message signal. Hence the VCO may be the kernel of an FM transceiver with dual purpose of transmit and receive in a 2-way time division duplexing (TDD) communication system.

[0050] A free running VCO is subject to frequency instability, such as temperature effects, component aging, power, and low frequency circuit instabilities among others. The communication link may be designed to compensate for the frequency instability of the VCO.

[0051] In one example, consider a two way communication system that is asymmetric in that the RV transceiver has to be small, inexpensive, low complexity, low power, but the ground transceiver is not subject to these constraints. The free running VCO is used in the RV and the GCS transceiver does the compensation for the free running remote transceiver based on the GCS free running VCO.

[0052] GCS compensation may be of the form that permits the GCS transceiver processing to maintain a model of the remote vehicle VCO as it changes with time. This trains on the uplink transmission, as used herein identified as the remote (RV) to ground (GCS) link. On downlink, as used herein identified as the ground (GCS) to remote (RV) link, the trained model is used to synthesize an FM modulated signal that has hopping and messaging at an offset in frequency that is closer to the IF frequency of the RV receiver.

[0053] Hence the remote would use the VCO with the known deterministic hopping signal (known to both RV and GCS) applied to down convert the downlink signal from the ground transceiver into the remote IF section of the receiver. The IF output may be a standard FM demodulator.

[0054] In the GCS, the VCO used may also be free running. However, the compensation added provides the warping of the VCO tune voltage that precisely results in the IF frequency offset as explained before. Hence both the remote and ground transceiver kernels are similar. Size, power, and cost may be of less concern for certain GCS implementations. As such, the receiver design may vary, while still being capable of communicating with the RV.

[0055] The training of the GCS compensation (model for the VCO tune voltage warping) may be based on the output of the power detector at the output of the IF filter used in the GCS receiver. By using tune voltage dithering and piecewise interpolation the warping function, the GCS adapts to changes of both the RV and GCS VCO.

[0056] An alternative to the RV IF is to use the VCO with reduced positive feedback as a narrowband filter. As the Q enhancement can be large, the additional filtering of the IF may not be necessary.

[0057] The RV +Q enhanced filter resonator (QER) may be modulated with the hopping signal such that it remains tightly synchronized to the downlink signal from the GCS transceiver.

[0058] Based on the orthogonality condition of the QER, the frequency of the VCO mode and the center frequency of the QER filter are precisely the same. Therefore the VCO / QER inthe RV transitions seamlessly between the transmit and receive modes of operation.

[0059] This VCO / QER may also be used in the GCS. Now the modelling will be to determine the relative compensating tuning voltage warping that is required. This essentially keeps the VCO / QER at the GCS behaving identically to the VCO / QER of the RV.

[0060] FRVT Implementation

[0061] The FRVT may be implemented as a simple free running VCO. The free running VCO may be modulated with a hopping frequency h and message content m (h+m) in transmit and h in receive whereE(h2) ≫ E(m2)where m may be M-ary but in a typical application is binary.

[0062] In the discussion below, examples and considerations are discussed with respect to specific implementations. It will be understood that other implementations may be based on different strategies and design considerations.

[0063] Dual Mode Transceiver VCO

[0064] In this example, consider a resonator which may be used with positive feedback in one of two modes:• A stable RF frequency Q-enhancing resonator (QER) in the left-hand jω plane, tunable in both frequency and bandwidth, or• In a self-oscillation mode (SOM) in the right-hand jω plane, tunable in both frequency and bandwidth.A modulation control input to this resonator is h0(t) which FM modulates the center frequency. The actual center frequency is h(t), which is different than h0(t) as the base resonator is free running and not phase or frequency locked. The difference ofΔh(t) = h(t) − h0(t)is a function of temperature, aging and other uncontrollable factors. The essence of the FRVT SA 400 is shown in FIG. 4, and includes VCO 20, IF filter 402, FM demodulator 404, BPF 406, fixed frequency synthesizer 408, TCXO 410, modulation processor / DAC 412, and antenna 414.Other components that may be used to implement a practical device, such as an LNA, initial BPF, power amplifier etc. will be described later.

[0065] The free running VCO may be modulated with the analog function of h+m or h. h and m may be generated by a digital processor with a DAC output and low pass filtered to a bandwidth of about 1 MHz or other suitable bandwidth.

[0066] Consider system A (SA) 400 as being the RV or remote transceiver and system B (SB) 500 as being the paired transceiver. The objective is that SA 400 is simple and inexpensive to implement with direct analog modulation of a VCO. SB 500 ‘locks’ to SA 400 and has more sophisticated processing. SA being based on direct analog VCO modulation may be frequency hopped at a fast rate and with a large tuning range. The utility is that this is achieved without extensive digital processing required for signal synchronization. The catch is that the simplicity of SA has to be offset with more complexity of SB.

[0067] A requirement is that h is known to both SA and SB, that SA and SB are properly calibrated, such as by an initial time synchronization, and that the modest TCXO of the SA may be accurately tracked and synchronized to the time of SB.

[0068] Note that a fixed frequency fifgeneration may be avoided if a Zl F down-conversion is used, provided that a quadrature mixer is also included. The FM demodulation would also become more cumbersome instead of using a simple phase locked loop (PLL). Further, we could have avoided the fifoffset in SA if we have the uplink and downlink carrier frequencies separated by fif. This is possible if the VCO in SA is at the frequency h and SB is at h+fif. This is the case if there is a different VCO in SA and SB. IF the VCOs in SA and SB are the same with identical FM modulation circuitry, the drift over time would beuB(w(h(t'))) - uA(h(t))which is more predictable.

[0069] In addition, there may be an oscillator, such as TCXO, that generates a local time base of t’. This may also be a free running oscillator and not tied to any reference time-base. Alternatively, it may be based on GPS time. The processing of the uplink signal may be based on t from SA.

[0070] Finally, there may be a power detector block attached to the output of the IF filter. This is for the PLC phase where the VCO of SB is frequency matched to the VCO of SA and the initial warping function of w() is determined.

[0071] The kernel of the dual mode transceiver is described by FIG. 5 and FIG. 6. In FIG. 5, VCO 20 in the system A is in self-oscillation mode (SOM) and generates a signal h(t) based on the h0(t) modulation input where t denotes the local time as produced by a local clock 502. This is transmitted to a second physically separated receiver and synchronized by a local clock 502 with time denoted by t’. The modulating signal of h0(.) is known to both system A and B. Receiver 22 of system B demodulates h(t) and jointly models Ah(t) and (t’-t), which is the difference in time between the separate local clocks of the RV and the GCS. This mode of operation is referred to as modeAB, or the mode of transmission from System A to System B. FIG. 6 illustrates modeBA.

[0072] The free running VCO is easily and directly modulated with the analog function of h+m or h. Both h and m are generated by some digital processor with a DAC output using a low pass filter bandwidth of about 1 MHz.

[0073] Consider system A (SA) as being the RV transceiver and system B (SB) as being the paired GCS transceiver. In this example, the objective is to make SA simple and inexpensive to implement with direct analog modulation of a VCO. SB ‘locks’ to SA and may have more sophisticated processing. With SA based on direct analog VCO modulation, it may be frequency hopped at a fast rate and with a large tuning range. This may be achieved without extensive digital processing required for signal synchronization. However, the simplicity of SA may be offset with more complexity of SB.

[0074] In a specific implementation, SA is comprised of a free running VCO that is frequency modulated by h+m, which is generated for the processor / DAC. As the VCO is not locked in a feedback loop it may be modulated fast and of significant depth. The direct modulation is fast, inexpensive and efficiently generates an FM signal. During transmit the VCO is directly connected to the antenna via the switches. In receive mode, the processor / DAC modulates the VCO with h. The down converted signal is processed in a conventional FM demodulator. TheSB transmits an FM signal that is offset by the IF frequency of fif. The IF filter provides the essential band pass filtering, centered on the 1 MHz bandwidth of the FM modulated m signal prior to the FM demodulator. Note the frequency of transmission is the VCO frequency. For receiving, the VCO is offset by a fixed frequency of the IF. As this is a fixed frequency the implementation may be easily integrated with a trivial increment in complexity. In addition, there is a TCXO which generates a local time base of t. This TCXO is free running and not tied to any reference time base.

[0075] The FRVT is based on TDD with alternating uplink and downlink transmissions as shown in FIG. 3a and FIG. 3b. The uplink may be defined as the transmission from SA to SB and the downlink as the transmission from SB to SA. The FM bandwidth is nominally 1 MHz such that the data rate is commensurate with this. The downlink and uplink frame durations may be variable depending on the data rate required in each direction. Normally the super-frame comprised of an uplink and a downlink frame should be short, such as less than 100 msec to avoid an increase in the latency in real time control data, resulting in sluggish response.

[0076] To account for the simplicity of SA, SB may be designe to do the more processor intensive task of precisely modelling and tracking the SB VCO. It may be assumed that the modelling parameters are known well enough to demodulate the FM portion of m. FIG. 7 shows a block diagram of GCS (SB) 500 which includes VCO 20, IF filter 402, FM demodulator 404, BPF 406, fixed frequency synthesizer 408, TCXO 410, modulation processor / DAC 412, and antenna 414.

[0077] To facilitate communications, both SA and SB must have a common point of reference. For example, it may be required that h is known to both SA and SB, that an initial time synchronization is done, and that the modest TCXO of the SA may be accurately tracked and synchronized to the time of SB.

[0078] In a TDD system, the network of system A and B will switch from modeAB to modeBA. In this mode, system B uses the model A / i(.) and the estimation (t-t’) to synthesize h(t) and transmit this to system A as shown in FIG. 5. The resonator mode is switched to QER mode and modulated by h0(t) such that it is equivalent to a filter that is matched to h(t). In this way, the coupling of system A and system B forms a time varying matched filter.

[0079] By alternating modeAB, seen in FIG. 5, and modeBA, seen in FIG. 6, the network may implement a TDD communications system. This allows system A to be implemented with an inexpensive asynchronous resonator with positive feedback and support almost coherent TDD communication, while system B may be a more elaborate GCS base system.

[0080] The discussion herein allows system A to be implemented as an inexpensive and low power consumption remote sensor that communicates with a base system, which may be more expensive and consume more power. In this system, h0(t) may be entirely arbitrary as long as it is within the capability of the modulated resonator in system A. It may, for example, be a continuous frequency hopping signal wherein the frequency range may be large, and the rate of change may be large, but the phase remains coherent. In this way a stealth signal may be generated that is difficult for a third party to detect, let alone effectively jam.

[0081] The data modulation in the basic scheme of FIG. 3 and FIG. 4 may be a simple AM modulation by amplitude scaling of the signal in both directions. The modulation may also be a constant phase FM modulation, which has the advantage of a constant modulus analog signal which is of significant advantage for efficient signal transmission, an advantage that is more difficult to achieve digitally. There may also be a mix of AM and FM that may be different for the modeAB and modeBA.

[0082] QER-SOM Mode and a Transceiver Kernel

[0083] There will now be described an example of a transceiver kernel that operates in a QER or SOM mode. The kernel starts with a frequency tunable resonator denoted as LC with a frequency control of f. The term LC generally refers to a resonant tank and is used because it is a likely integrated implementation of the resonator for the QER and SOM described herein, although other suitable resonators may also be used, such as a SAW. For example, the LC component may be any system that stores energy and has a frequency response with a set of natural resonances, and may not have a single resonance. A feedback loop may be provided around LC such that, with a moderate level of active feedback around LC at the appropriate phase shift, the behavior of the system, LC and the active feedback path, will have characteristics of a single dominant pole resonator of moderate Q. This system is denoted as LCaf1002.

[0084] The single dominant pole created from the active feedback may be tuned in center frequency by a control f. The gain of the active feedback may be controlled by an input g. If g is increased, the bandwidth of the system decreases providing a useful bandpass filter response. This is the QER or Q enhanced resonator mode of LCaf1002 as seen in FIG. 8. If the orthogonality condition (OC) is satisfied, then the center frequency of the bandpass response of the QER will remain constant with changes in g. Hence, ideally, the Q of the LCaf is a function of g only and the center frequency is a function of the f input only. While this may not be the case in all implementations, it will be assumed for this discussion that the center frequency of the LCaf is a monotonically increasing function off and that the Q of the LCaf is a monotonically increasing function of g.

[0085] A further increase in g and the mode of LCaf1002 will change to a self-oscillation mode or SOM as shown in FIG. 9, In this diagram, there is the added complexity of a synthesizer, but if this is a fixed frequency, it may have a relatively low complexity to implement. VCOd and VCOg frequency modulations may differ by only a small m(t) to facilitate the modelling of g(). The OC may be used to assure that the SOM oscillation frequency will be the center frequency of the QER filter. In the SOM mode of the LCaf, g is constant, and held close to the threshold value of the SOM mode. This may be determined directly with a detector applied to the output of the LCaf as g may be increased for an FCaf with no input until the detector ‘goes high’, i.e., shows active RF signal power at the output of the FCaf. Over a wider frequency range these g values may be described as a function of f even though the OC may be satisfied locally. Hence the calibration may include determining g required for the onset of SOM as a function of f. This is stored in a LUT such that a change in the f input will trigger an interpolation of the LUT of g. As such there is only one input which is the frequency control f. Consequently, the LCaf I SOM behaves as a VCO device with a single f input. Finally, it may be noted that the requirement of g being minimal for SOM is that the OC is satisfied and that the frequency of SOM is then equivalent to the center frequency of the QER mode at moderate to high Q.

[0086] The center frequency of the LCaf in either SOM or QER mode may be expressed as fc = «( / )where u(f) is a function specific to the physical copy of the LCaf. u(f) is ephemeral, requiring continuous modelling as the components of the LCaf are subject to aging, temperature, changingsupply voltages, etc.

[0087] VCO as the Q Enhanced Resonator

[0088] The VCO may be switched from the SOM oscillator mode to a Q enhanced resonator (QER). In the QER mode, the VCO is not acting as a resonator with an active stable feedback filter that modifies the bandwidth and provides a gain enhancement via processing that may modify the amplitude of the feedback signal. At the system level, an example of a VCO in the QER feedback mode is shown in FIG. 10, with VCO 20 switched into the loop with QER 1202, FM demodulator 404, and processor 412.

[0089] Frequency Correction

[0090] A suitable way of tracking the RV frequency during run time may be to use the detector output of the IF filter and ensure that the detector value is maximum for all the frequencies. A problem is that the detector only gives a power output of the IF component.

[0091] The corrector function of gc() is applied to the voltage input of the VCOg. To optimize the detector output for all the frequencies, dithering is used in this example. In a new frame, the function gc() is either incremented up or down in value. If the new value gives a better detector value, then the new value is used. Conversely, the old value is retained.

[0092] As there is only the detector output and the VCO curves are nonlinear, it may be useful to start with a piecewise linear correction function where we assume an input voltage of 0<v<1 that is input.

[0093] Model of a VCO

[0094] It may be assumed that the VCOd and VCOgmay be tuned over the frequency range of 3 to 4 GHz. However the tuning curve will drift with time and is not linear. Start with a simple model for VCOdfrequency output depending of the input argument of the bias voltage v in the range of 0 to 1.f = gM = 3 + vLikewise the model of the GCS VCO is a function of voltage asf = 9gv = 4.5 + vNow a correction to v may be described as0CO) = 1 * vwhere the 1 is the gain coefficient. We assume this additional processing such thatf = 9g 9c v )The functions for the RV models are done with piecewise linear curve fit of the input voltage which ranges from 0 to 1.

[0095] For a code, gc() is given as a linear piecewise model with coefficients stored in the LUT. Initially the values of gcis an array of (M,) where M is the number of value pairs. Assume a uniform array from 0 to 1 asV01 = np. linspace(0,l, M)The gc values correspond to these voltage values. In the processing the input voltage will be 0 to 1 volt as shown in FIG. 11. In the VCO hardware this will be scaled over the range of the VCO tuning port, but this is considered from the algorithm to be part of the VCO. The linearly interpolated value of gcis added to the input voltage.

[0096] Pre-Launch Condition (PLC) generation of w()

[0097] There will now be discussed an example of a PLC example. Initially m(t)=0 and h(t) is static. The 10 MHz detector is used to determine w(h(t)) for each static frequency of the VCO. Then the 1 MHz detector is used for a refined value of w(). Next the w(t) is smoothed with an interpolating function which is stored in the LUT.

[0098] The hopping range is 1 GHz and accuracy is preferably within a target 100kHz. Hence an accuracy of 100 ppm relative to the hopping range. Relative to the absolute frequency, an accuracy of 20 ppm is preferred. Note that the bandwidth of 1 MHz will not provide 100 kHz resolution the centering of the frequency.uB(w(h)) +fif2- UiC / i)The initial TCXO may be trained to within 1 ppm. 1 ppm is 1.5 MHz offset at the start. However, this is offset by the VCO training of w(). The frequency may be placed within 100kHz with dithering of w(h) that determines the band edges of the 1 MHz BPF and this will provide theinitial VCO characterization to within 10 MHz. Then the 1 MHz detector may be used as a refined search.

[0099] The function h(t) is a pseudo-random process with an autocorrelation function that has a kernel in delay time that spans about 1 μsec. The conventional locking loop has an early-late gate structure that looks at the correlation of the signal with h(t-td).

[0100] In the discussion below, it is assumed that the warping function is accurate, and the delay is sought. The function uA(h(t-td) is not known, and the principle is a PSD kernel width of 1 MHz. Hence, it decorrelates in 1 μsec. Early / late gate on output of PLL. h is 109Hz and we are 105Hz, so we must have timing accuracy of 100ppm. Another way of stating it is that we have a frequency slew rate of 109x106= 1014Hz / sec. Can’t be off by more than 105Hz which is then 1 nsec. Changes at 1 nsec in 0.01 seconds. Assuming velocity does not change, 100 MHz clock gives 10 nsec resolution.

[0101] Simulation of the Detector Output

[0102] There will now be described a simulation of the detector output. The IF frequency is ff which initially is assumed to be 1.5 GHz. The error frequency is given asferr 9d(v) fifThe detector output is modelled as1^det 7 2BIFis the bandwidth of the IF which is initially.01 GHz but will eventually be.001 GHz. For now assume a simple single pole IF filter of one dominant resonator.

[0103] In an FM radio system, a tunable bandpass filter is needed for the receiver implementation and a VCO is needed for frequency translation as well as angle modulated signal generation. The LCaf may provide both with the g controlling the mode change between QER and SOM (VCO mode).

[0104] Specifically for FM angle modulation transmission, the SOM may be used to generatethe signal directly. The QER may be used as a bandpass filter or as a time varying matched filter that may be trained on the FM angle modulation.

[0105] For a TDD protocol, LCaf may be switched between the SOM mode used in transmit and the QER mode used in receive.

[0106] The pair of LCaf devices may be similar circuits and manufactured from equivalent components but will not be perfectly matched. Hence the {g,f} for a given behavior in one system will not generate precisely the same behavior in another system. However, one system may be trained such that its behavior mimics the other system. Due to OC, it is only necessary to mimic the behavior of the center frequency it is only necessary to warp the frequency control of f of one system to generate the same response in the other system.

[0107] Propose a notation that works to describe this. Let Vi be the voltage of the frequency control into system 1 that produces a center frequency of fi. We assume the OC such that fi is the same as the center frequency of the QER response and the VCO frequency. Likewise, f2is the same center frequency of the QER response and the VCO frequency for system2 which is generated by the frequency control voltage of v2. The warping may be described as follows. Define ui() as a function that maps Vi to h and u2() that likewise maps v2to f2asfl = Ui(Vi)f2= u2(v2)It is desired to have fi = f2which may be achieved by taking the voltage Vi, that produces fi in system 1 as the input voltage to system 2 but we warp Vi with a warping function w() such that when applied as the input voltage to system 2 we get the frequency fi. That is fi = u2[w(vi)].

[0108] In this system, the actual frequency remains unmeasured, leading to:Ui(Vi) = u2[w(vi)]In the application, the functions u1 and u2 remain unknown. However, fora given applied voltage Vi the warping function w() may be adjusted such that ui(vi) = u2[w(vi)] is satisfied. For these purposes, it may be assumed that the warping function is available. How w() is determined will be described below.

[0109] Initially assume that w() is known such that ui(v) = u2(w(v)) for a common tuningvoltage signal v. It is not that relevant in this application exactly what the tuning curve of ui() is. It is not necessary to target an application where the output transmitted frequency is precisely regulated, such as may be found in civilian applications.

[0110] Consider the FM transceiver link shown in FIG. 12 that includes a LCafi 1402, LCaf2 1404, and FM demodulator 404.

[0111] LCafi 1402 in SOM transmits an uplink signal with a voltage of Vi = h+m. Here h denotes the base frequency bias which will later bet the hopping, h is known to both the transceivers, m’ is the receiver estimate of the FM transmitted message m. The receiver QER bandpass filters the receive signal and the FM demodulator 404 extract m’. The FM demodulation will be based on some form of PLL.

[0112] The path may be reversed such that LCaf2 1404 switches to SOM and transmits the FM signal based on vi=h+m and LCafi 1402 changes mode to QER and then forms the receiver as shown in FIG. 13.

[0113] By alternating the configurations shown in FIG. 12 and FIG. 13, the TDD protocol is implemented. The dwell time in each of the two modes may be varied depending on the data flow just as a conversation between two individuals. It may be a fixed allocation of TDD dwell time or variable. This is in the prior art and will not be considered further here.

[0114] Next consider h as being a hopping function of time t. Consider first a slow hopping function of h(t). The hopping would generate a frequency of ui(h(t)) that may span 1 GHz or more. If h(t) is slow then the phase-locked loop (PLL) will have no issue with keeping up and demodulating the combined signal of h+m. It may be shown analytically and by simulation (see prior document) that if LCaf2 with v2=h and LCafi with vi=h+m that the frequency modulated signal will flow through the QER uninhibited. That is as long as the bandwidth of the FM modulated m(t) fits within the instantaneous bandwidth. Typically, it may be assumed that |m|«(variation of h) such that the QER is only misaligned with respect to m(t).

[0115] Preferably, h(t) and m(t) are continuous functions of time, which is different than conventional hopping where h(t) is a discontinuous stepped function nd allows phase coherence to be retained. In conventional hopping, h sets a new frequency, and the source must lock upagain as well as the receiver. This takes time. Conversely, in this discussion, h may remain phase coherent.

[0116] The PLL which synchronizes to the output of the QER locks onto the carrier given by u1(h+m). Note that with varying h the PLL no longer tracks m directly, but h+m and that m is a small component of h. However, h is known and therefore, even if the amplitude of h emanating from the PLL is not known it is possible to extract m(t).

[0117] Also, for slow h the time synchronization is not an issue as the PLL tracks h. This tracked h may be correlated with the known h and then the timing error extracted from which LCaq2 may be synchronized with LCaqi.

[0118] However, let us now move to a faster variation in h(t). This poses a problem for the PLL. But if it did not then the link could be easily jammed by a reactive jammer. FIG. 14 shows a basic PLL 1600 with VCO 20 and low pass filter 1602. Consider this basic PLL 1600 that could be used for the FM demodulation:u^h + m) = uvco(v)such thatv= uvco-1(uiC / i + m))where uvco-1is the inverse function of uvco( ). Clearly if the VCO of the PLL is a SOM calibrated with respect to LCafithen the uvco-1cancels) and v=h+m.

[0119] Suppose we could train LCaf2 and LCaf3 (or a VCO) from LCafi to run in SOM mode. Then we may get an undistorted modulation voltage of h+m. But the PLL will have trouble tracking the wide ranging frequency u1(h+m) through the 1 GHz excursion with numerous interferes. A powerful jammer will essentially trap the PLL. Instead we do the implementation shown in FIG. 15 that includes an additional input of uvco~

[0120] It is this offset that the jammer does not have and therefore cannot track the signal through any other interference. This also means that the h(t) may be independent of m(t) in that it may have a broader bandwidth. Take the VCO and offset with an IF of 1.5 GHz and then a 10 Hz channel.

[0121] Initial and Run-Time Calibration

[0122] The RV may use a free running fast frequency modulated VCO. By not locking the VCO, it may be frequency modulated faster in an analog fashion. In one example, the goal may be a 1 GHz tuning range and 1 MHz modulation. This may be achievable as COTS VCOs may have 30% and may have a modulation bandwidth that is 0.1% of center frequency. With an unlocked VCO, the RV transceiver becomes very inexpensive. However, the GCS becomes more complex as the downlink must be ‘frequency locked’ to the RV with the instantaneous bandwidth on the order of 0.1% of the center frequency. In this example, the RV has a free running clock that sets up time t, with t=0 reset for pre-launch. GCS has clock which is t’ and is calibrated such that t=Cit’+b where Ct and b are model constants determined by the time synchronization of the GCS. Again, no synchronization in the RV itself simplifies this circuit.

[0123] The objective in the present discussion is to step through the calibration and tracking process. Thereby scope out what is needed for the GCS processing.

[0124] Denote the pre-launch calibration and configuration procedure as PLC. The RV may be completely asynchronous with a free running VCO. This simplifies the RV, but to do that there is more complexity in the GCS.

[0125] An example of a PLC is shown in FIG. 16. The PLC setup with the MCUs 1802 of RV 100 and GCS 104 connected with WiFi or some form of connector 1804 such as a simple optical fiber bi-directional UART and clock. This may be convenient when the RV is sitting on a launch pad near the GCS. The PLC will disconnect when the RV takes off. The PAs could be scaled down in power for the PLC. WFi link insecure, a PLC cable connection may be used configuration data is sensitive.

[0126] The connection between the two MCU’s facilitates the coordination of the functions during the PLC.

[0127] A possible receiver that uses a hopping band of 3-4 GHz and a 10 MHz IF band is shown in FIG 17 and includes a 3-4GHz 10dB LNA 1902, 3-4GHz filter 1904, 1.5GHz SAW 1906, d / dt filter 1908, and ADC 1910. Note that the IF frequency should be high enough that the tuning bandwidth of the VCO is less than an octave. An IF of 1.5 GHz works as the VCO covers 1.5 to 2.5 GHz. Note we need the high IF frequency such that there is no image band issue. TheIF frequency should be larger than the hopping range.

[0128] Three factors to consider are:1. Where the VCOd is offset by 1.5 GHz for receiver operation, it may cover the bands of 1.5 - 2.5 GHz and 3-4 GHz. A tuning range of 1.5 GHz to 4 GHz may not be practical.2. A 1.5 GHz SAW with a bandwidth of 1 MHz will require a pole Q of several thousand.Preferably, we should aim for pole Q’s of no more than 500. Hence the SAW bandwidth may be more on the order of 10 MHz.3. The actual frequency of VCOd should be known to a sufficient degree to the GCS to hit the IF bandwidth. If the differentiator is used, then the slope side of the d / dtfilter should be about 1 MHz.

[0129] A solution to the third factor will be described later as part of the GCS processing. A solution to the first and second factors may be to use zero intermediate frequency (ZIF) with quadrature processing shown in FIG. 18, using two LPFs 2002 and ADCs 1910. However, this is effectively a GCS which is less desirable. However, as the bandwidth may be made small, for example 100 kHz, this is a possibility and then using QPSK instead of FM which may still be approximately constant modulus. But also, the cresting factor of the GCS power amplifier is less of an issue.

[0130] A solution to the first factor may be to offset VCOd by the IF frequency. Hence it is easy for the GCS to model VCOd which runs with h0(t) for receive and h0(t)+m(t) for transmit.

[0131] FM Modulation

[0132] In some implementations, FM modulation may be used to transmit a message signal. The FM mode was shown in FIG. 5 for the modeAB. The SOM of system A is modulated by h0(t)+m(t) where m(t) represents a message signal that is initially unknown to system B. System B jointly estimates m(t) in addition to the SOM model of System A and the time difference.

[0133] A scheme that may be used to address the first and second factors is shown in FIG.19. The signal received by antenna 414 is amplified slightly by LNA 1902 to manage NF (may not be necessary) then filtered to the 3-4 GHz hopping band by filter 1904. Then the signal isdown converted to the IF frequency of Nf0by a 1,5-2.5GHz BPF 2102. This is a fixed frequency generated by a synthesizer. As the synthesizer is fixed frequency, we should be able to find one that is inexpensive. One choice is to have Nf0of around the GPS frequency of 1.575 GHz with a 10 MHz bandwidth. We then have a second IF of frequency f0with a narrow bandwidth on the order of 500 kHz. Then a phase locked loop (PLL) to demodulate the FM. The PLL is simple as the signal bandwidth is reduced to about 500 kHz.

[0134] Note that for uplink transmit mode the VCOd is modulated with h0(t)+m(t). On receive, it is modulated with just h0(t). During the receive mode the VCOd is offset by Nf0The input frequency is the VCOgmodulated by g(h0(t)) (implying that it is modelled based on VCOd with the input of h0(t). Hence the IF frequency of Nf0should be accurately matched. The output of the IF SAW is then down converted to a second lower IF such that the bandwidth may be reduced by at least a factor of 10.

[0135] On the GCS side, a similar transceiver architecture, as shown in FIG. 20. However, the uplink modulation bandwidth may be higher such that the second IF is not strictly needed. However, it may still be used it as the bandwidth is 1 MHz. A second IF may be used for reducing the bandwidth from 10 - 20 MHz down to 1 MHz.

[0136] As there is now an IF frequency, the VCOd and VCOgwill need to be in a sufficiently accurate pseudo frequency lock, and the model function g() will preferably be updated continuously to allow the frequency lock to be maintained as conditions change. Consider the uplink transmission where the PLL is decoding m(t). However, this does not provide any input as to the correction of the VCO model. VCOdand VCOgmay be offset slightly in frequency and still be frequency misaligned for the narrow IF filter bandwidth. To get an observable useful for the alignment, the detector output of the BPF may be used. m(t) will have binary values of + / -1 (scaled by some constant). The +1 and -1 modulation output of the IF should have the same amplitude. However, if there is a misalignment in g() then the detector values for m = +1 and m = -1 will be offset. The direction of the offset is an indication of which way to nudge g(t) for the specific hopping frequency. Hence for every value of h0(t) which corresponds to a frequency on the VCO tuning we get a different nudge value. Then g() is updated.

[0137] A simple first order simulation of this observable is described below. In this simulation, the IF filter is represented by a second order bi-quad. The frequency is changed representing achange from m = -1 to m = +1 at the normalized time of t = 500. There are three simulations with the input frequency too low, correct and too high as seen in FIG. 21, FIG. 22, and FIG. 23. It is evident that we may sample the detector output to get a measure of the misalignment. Note that in reality the signals will of course be noisier, and it is necessary to average many data samples before committing to an update of g().

[0138] Each detector observation is a set {h0(t),m(t), detector output}. The correction only works if we have the bit value of m(t) jointly estimated. This is typical of receiver synchronization schemes in that there must be some structure of the signal that may be used to correlate the synchronization correction to. The scheme must be tolerant of errors in m(t). However, we may also have an initial framing sequence to the uplink transmission as this is not excessive overhead. In this case the first say 100 bits of m(t) are a known code word that may be used for the update to g().

[0139] As the correction to g() at one frequency is correlated with a change over a range of frequencies then the correction update uses an exponential shaped kernel with a width that is commensurate with things like the frequency response of the bias circuit.

[0140] FRVT vs Conventional Analog FM

[0141] In a conventional analog FM, the transmitter has a modulation frequency of h+m where h is the carrier offset frequency and m is the message. The receiver has an approximate knowledge of the carrier offset given by h’ and an if frequency of fif. After receiver down conversion the frequency of (h’-h) + fif+ m passes through the IF filter and then onto the FM demodulator which may be a PLL or some form of LC network.

[0142] In the FRV, h may be continuously variable over a wide frequency range of, say 1 GHz, and fast change of, say, a 1 μsec time constant. The receiver synchronizes time with the transmitter such that h is precisely known. However, there will still be a small frequency error such that of (h’-h) + fif+ m is the frequency modulation passed onto the FM demodulator.

[0143] So the fundamental difference is that h in the FRV is hopping rapidly over a broad frequency interval and is an independent function relative to m. That is h may have faster variation than m or slower or variable, h and m are independent continuous time modulations.So the difference may be described as that of hopping h rather than static h in comparison with the analog FM.

[0144] Next, consider the digital hopping FM where the carrier is hopped to a specific frequency and settles for a short dwell time before hopping to the next frequency. During the dwell time a short message frame is sent. But this scheme is different. We have h+m+s where s is the settling transient of the PLL which locks / synthesizes h. s is random and spoils the phase coherence, effectively interrupting m at every discrete frequency hop event. So we could implement h+m with a DDS or high speed AWG, although this may be more expensive when implemented in the remote RV end (system A or SA).

[0145] With that, the implementation may be based on a free running VCO frequency modulated by h+m in the transmit mode of a communication system wherein h and m are continuous time modulation functions that are mutually independent.

[0146] Pairing h’ to h

[0147] There will now be described an example in which h’ is paired h. In SA, h is the deterministic hopping signal and m is only known to SA and not SB; h+m is the signal synthesized by the SA processing that is the tuning input into the free running VCO of SA. The actual frequency is uA(h+m) where uA() represents the frequency tuning function of the VCO which varies with time. The LO frequency generated by SB is uB(h)+fifwhere fifis generated by a frequency translation in the SB. Here uB() is the VCO tuning function of the SB. The VCO of SA and SB may be the same circuit architecture with equivalently valued components but they will not be precisely the same and therefore uAQ = = uBQ. Furthermore, uAQ will drift in time differently than uB() primarily as the ambient factors and aging are different. Therefore, uAQ has to be paired or matched with uB() by some adaptive process.

[0148] Assume that the PLC (pre-launch calibration and configuration) has been done such that initially this pairing is accurate. The method of doing this is to have a ‘voltage warping function’ denoted by w() that satisfies the frequency matching relationship of u4( / i(t)) = uB(w( / i(t))). In PLC the time may be precisely synchronized, and h(t) is shared between SA and SB.

[0149] Assuming uAQ and uB() are monotonic functions determining the initial warping function ofw() is straightforward.

[0150] For the adaptive tracking we have the frequency ofuB(w(h(t'y)} + fif- uA(h(t) + m(t))This passes through the IF filter which is in two stages of a 10 MHz followed by a 1 MHz.

[0151] Propagation Delay and Run Time Synchronization

[0152] In order to accommodate the propagation delay between the RV and the GCS, consider a delay of about 1 nsec / foot, such that a delay of up to 104 nsec or 10 psec is possible, which is several bit periods and much longer than the eventual time constant of h0(t). Let the propagation delay be Dp. For the uplink synchronization, the GCS time t’ is aligned with t-Dpwhere t is the current time at the RV. This is illustrated in FIG. 24. Dpalso includes the delay through the RV transmitter and GCS receiver.

[0153] As the PLC is done with high SNR, there is no issue in the initial time synchronization. The TCXOs of the RV and GCS after this initial synchronization will only change by say 0.1 ppm (TCXO in RV is still warming up and stabilizing). A typical flight of 1000 seconds still means that the number of bit intervals separation may change by 103 + 6 - 7 = 100. Hence run time synchronization is necessary. However, that is not an issue as the uplink frame has structure and there would be an identifiable start code. Say the interval between start code of two successive frames is 30 msec measured with RV time t, then that is a precise number of ticks of the RVTCXO. The GCS aligns the start of the frame with a clock edge of the GCS TCXO and then counts the number of clock cycles to the next frame which precisely sets the difference in t and t’ such that we can map t-t - Dp.

[0154] For the downlink, it may be necessary to generate h0(t+Dp) such that the uplink transmission will be aligned with h0(t) at the RV. The RV is as simple as possible which therefore excludes any clock recovery processing in the RV. Also, it may be beneficial to have the RV VCO operate continuously with h0(t) (modulated with m(t)) without further local synchronization. This is illustrated in FIG. 25.

[0155] For this time alignment, the detector output feedback of the RV receiver may be used, which is sent as telemetry samples back to the GCS. In this way the GCS can dither Dpslightly and observe the detector output. Note that we put one dither cycle in every downlink epoch with three detector measurements. Hence will have a measure of the alignment every 30 msec. Say that we have a stabilization loop with a bandwidth of 1 Hz. The typical Allen deviation for a TCXO is about 10’11for a T of 1 second as shown in FIG. 44.

[0156] By way of example, for a symbol period of 1 μsec, we would be misaligned by about 20 ppm which is negligible. The bigger factor is the uncertainty in the actual Dp. The RV can stop, accelerate, maneuver and so on within 1 second. We can have an unmodelled deviation of a maximum of 100 feet within a second interval which is 100 nsec of uncertainty in Dp. This is 10% of the symbol period which is much more significant. However, this uncertainty may be reduced with a Kalman filter built into the processing. The Kalman filter would model the range (Dp) based on inputs from the controller that are available (possible GPS, CV, IMU etc.). Steady pilot flight would significantly reduce this uncertainty. However, even if a rudimentary random walk model for Dpis assumed then the tracking of delay and the amount of advancement of the hopping sequence is not an issue. Complexity increases with a wider hopping and message bandwidth.

[0157] During downlink, data detection samples are taken and stored by the RV. These are sent uplink to the GCS for the estimation of the timing loop to control d. Note that d may also be estimated from a variety of other sensors that may be present (GPS, CV, RSSI etc.) The faster the h0(t) modulation the more accurate d has to be estimated.

[0158] Note that as the TCXO is far more stable than the actual changes in Dpdue to flight dynamics, we may use the TCXO and changes in Dpfrom the RV detector sample synchronization to get accurate estimates of RV radial distance and velocity.

[0159] GCS PLL

[0160] Consider the linear model of the PLL of the GCS. There is a step change in input frequency due to m(t) which is represented as the phase ramp of Additionally, there is thehopping frequency slope which will be represented by. The VCOghas a Laplace ofs3HycO^S)s(s +. b ib\)here kvcois the gain of the VCO which is the voltage to frequency slope. The bias circuit of the VCO resonator varactor is represented by a first order LPF with a bandwidth of b. The RF LPF that follows the mixer has a transfer function ofThe loop filter has a transfer function ofkf(s + bf}n,r<^ =Asf)Additionally we add in the offset frequency slope to the VCO input. Therefore the error of e(t), seen in FIG. 26 should converge to zero in the steady state.

[0161] Note that the effective bias input in terms of phase islU L _ bb\ _ hd / s \ _ 1 / / id\s3\ s + bb) s3\s + bb) s2\s + bb)which is equivalent to a steady state phase ramp or an offset frequency of—. This is equivalentbbto the delay of the bias. The delay will be compensated by the control loop adjusting the delay of the applied modulation. Another possibility is to put a similar low pass filter in front of the PLL which is like the VCO bias filter. The input phase of m(t) and h(t) is shown in FIG. 27.

[0162] In summary, the VCO bias may require an LPF with bandwidth of at least 10x the PLL loop bandwidth to avoid closed loop poles that are oscillatory that could become unstable. The input signal amplitude into the PLL may be controlled to avoid PLL gain modulation, such as by using an AGC prior to loop.

[0163] Likewise in modeBA, system B uses a message signal of m(t) to send back to system A which is now in QER. A key point is that this transmission from system B is h(t)+m(t) in which the larger h(t) component is coherent with the QER matched filter at the present time, which is facilitated where m(t) is correctly demodulated / decoded.

[0164] There is a variety of opportunities here with concepts still being developed. For instance we can have a pair of resonators that can operate in SOM / QER modes and thenbecome synchronized by a form of mutual training. For instance, resonator 1 in SOM can train resonator 2 in QER. Then these may be used to further the demodulation of h(t)+m(t). Suppose that m(t) takes on binary values of x and -x. Then we can have QERi modulated to h(t)+x and QER2to h(t)-x as a pair of time varying matched filters.

[0165] Additionally, there is a scheme where the system A resonator stays in SOM. In modeAB it is modulated with h0(t)+m(t) and in modeBA it is modulated with only h0(t) as m(t) is unknown to system A. The SOM is used as and local oscillator for down conversion of the system signal for a conventional ZIF demodulation, but this may require quadrature downconversion. The other option is that the SOM is modulated with h0(t)+hiFfollowed by a conventional superheterodyne with an IF frequency of hiF. Note that in the modeAB, system B will train on h0(t) and knows hiFused in system A and therefore can correct for any frequency offset in the SOM.

[0166] SA and SB as a PLL model

[0167] FIG. 29 depicts systemA (SA) 400, which may be the RV and systemB (SB) 500 which may be the GCS. SA 400 has time t based on TCXO 410 and has free running VCOA20B. SB 300 has time t’ based on its TCXO 410 and has free running VCOB 20B. The delay time between the SA and SB will vary as SA moves away from SB. Also the tuning curve of VCOA will drift from VCOB. Instead of joint estimation of the changing delay time and tuning curve we consider a separate tracking. Add in a short calibration epoch for VCO training in the TDD and a PLL during receive mode of SA and SB to track the delay.

[0168] Initially assume VCOA 20A and VCOB20B have identical tuning curves and t=t’. The transceiver concept is shown in FIG. 29 where VCOA is modulated with h+m and VCOB is modulated with h. Then we can use a quadrature ZIF to extract m.

[0169] In one example, assume a system that is capable of a continuous frequency hopping slew rate of 100 MHz / psec and in which the hopping range is 1 GHz. In this example, the modulation bandwidth is nominally 1 MHz based on the FPV data rates required. Therefore, we have a tolerance of the frequency mismatch of about 100 kHz, otherwise the signal m(t) will not pass through the LPF filter 2002 in FIG. 29. Therefore, the warping function applied to the tuning voltage of VCOB must match VCOA within 100 kHz. Say the tuning voltage range is 0 to 10 voltsfor the 1 GHz range then we must have 1 mV accuracy in the tuning voltage which is possible. Another way of stating this is that we must have 100ppm accuracy relative to the full scale of the DAC driving the VCO which is possible with a 14-bit DAC. The frequency slew rate is 1014Hz / sec and we need to maintain an accuracy of 100kHz so that the error in t-t’ cannot exceed 1 nsec.

[0170] Suppose we synchronize in PLC phase (pre-launch calibration) such that t=t’. Then during flight, the delay is going to add at least 20 nsec / sec due to RV flight and the TCXO of the SA and SB are going to drift. Based on typical Allen deviation of a TCXO this is less than 1 nsec / sec so negligible in comparison with the flight dynamics. Also the VCOA is not burned in and stabilized and will have a higher Allen deviation than VCOB.

[0171] One method for using FIG. 29 is that the VCOs are trained with respect to the TCXO. A possible method is shown in FIG. 30 where we have VCO calibration epochs 3202 squished between the transmit epochs 3204 and receive epochs 3206. A VCO calibration epoch 3202 may be as short as 100 psec, calibrating a subset of the frequencies in the VCO tuning range. Say we have 10 psec for each frequency. Note PLL bandwidth is 1 MHz and the existing LUT can be assumed as accurate. Hence the PLL essentially must lock up in phase and the LUT error will be on the order of several kHz at most.

[0172] We only need to do a small set of frequencies as the spline interpolation of the otherwise smooth VCO curve gives sufficient accuracy. With 100 psec we can do say 5 frequencies in each VCO cal epoch. Choosing frequencies to be multiples of TCXO frequency simplifies.

[0173] The calibration may be implemented by using switches around the VCOA 20A to tie it into a PLL that locks on a reference frequency generated by the TCXOA 410A via a divide by N 3302 as shown in FIG. 31. In this way we can get the precise VCOA voltage within 100ppm required for the VCOA generating the frequency of N times the TCXOA. The PLL lock is fast as it only has to adjust to the change in the value from LUT 3304 from the previous calibration. Hence this will be very close.

[0174] FIG. 32 shows the LUT of the samples generated at N times the TCXOA frequency. Interpolation of the LUT is done say with a linear, quadratic or cubic spline. If there are sufficient points in the LUT and that the tuning curve is smooth, then linear interpolation may be sufficientto stay within the 100 ppm frequency error requirement. However, it may not be practical update all the frequency points in each LUT calibration interval as there is not enough time, while But a quadratic or cubic interpolation requires more processing.

[0175] The assumption is that with continual LUT calibration we can maintain the frequency error within the 100ppm for the VCOA. The same process will be applied to VCOB of SB. Note that this form of calibration is done in modern spectrum analyzers and VNA’s such that a fast continuous frequency sweep is possible. 100 ppm accuracy is possible. Also, in PLC the frequency of TCXOA and TCXOB are calibrated relative to a frequency reference such that the calibrated absolute frequency difference between TCXOA and TCXOB will be within 0.5 ppm.

[0176] However, TCXOA will drift relative to TCXOB. The Allen deviation is 10’10at 0.1 sec interval for a low cost TCXO. The variance is then 10’20and over the 1040.1 sec intervals of a 20-minute flight we have a deviation of 10 nsec which is too much. Continual correction is required. However, there are ample observables available to maintain this below 1 nsec absolute deviation. Moving forward, we will assume that the calibration of VCOA and VCOB is sufficient to be within the 100 ppm such that we can focus on the time synchronization and assume that the VCO tuning mismatch is negligible.

[0177] In summary, the TCXOA sets t and TCXOB sets t’. Assume that VCOA generates h(t)+m(t) during uplink transmit. In SB, time synchronization is done by adjusting g() and td where t-td-g(t’)=O, where g() is some calibration function (offset bias or affine is sufficient) and td is the propagation delay. Now in SB, tdand g(t’) are linearly dependent and it may be difficult to estimate separately regardless of SB processing. Also, tdvaries significantly with up to 20 nsec / sec which dominates so it would make sense to ignore updates in g(t’) and just attribute the timing variations to td. Then SB can transmit based on g(t’+td) such that we advance the time by td and at SA using h(t) for demodulation will then be in sync. However, the receive and transmit paths in SA are not identical. It is therefore better that both SA and SB estimate td where tdis applied to the receive function. Therefore, SA transmits h(t) and SB demodulates with h(g(t’)-tdB), where tdB is the estimate of td made by SB during uplink transmission. In this case we don’t distinguish between g(t’) and tdB such that g(t’) is frozen from PLC. Likewise, SB transmits h(g(t’)) and SA demodulates with h(t-tdA), where tdA is the estimate of td made by SA during downlink transmission.

[0178] The tracking of tdB may be based on a PLL where VCOB 20B is now locked to VCOA 20A h+m modulation as shown in FIG. 33. No modelling required but now the modulation input to the VCOB will have the precise delay and VCO model difference built in. The h+m output may then be further demodulated to produce m.

[0179] A problem with the PLL in FIG. 31 is that the large fast frequency variation of h+m must be tracked. Another issue is that if there are bands of high interference then the PLL will not track through the band of interference as h is changing.

[0180] We can aid by adding in the estimate of h as shown in FIG. 34. If the estimate of h is perfect, then the PLL only must adjust to m. Then also m is directly demodulated. The problem is that h must be accurately estimated which is done by an ancillary loop 3402. This loop is facilitated by the knowledge of h() such that the model generates w() and g() and h(w(g(t’)))should match the h modulation of the VCOA.

[0181] The problem of the PLL is that it may cycle slip which causes a symbol loss in the m(t) demodulation. It may lose lock completely and require a reacquisition. The challenge is therefore to maintain a sufficiently accurate model. Suppose that the model of h is accurate such that the PLL only has to contend with the noise in the 1 MHz band of m. The normal PLL SNR calculations for AWGN and threshold effect may be used as guides here.

[0182] In some examples, it may be useful to model the PLL as a Kalman Filter (KF) as then we can start making the loop more complex with higher order and with the ancillary model loop also included. The objective is to determine the performance limit of the PLL with the modelling uncertainties. The advantage is that the PLL can accommodate a model of h that is inaccurate and still extract m. The disadvantage is that the PLL easily looses lock if the model of h is inaccurate.

[0183] T o start, consider modelling a first order PLL using a discrete time KF updated every 100nsec as the time constant of m(t) and h(t) is 1 μsec. Use the Riccati equation to determine the steady state phase deviation. Consider discrete state KF which makes sense for the time delay state oft’ which comes from a TCVX and counter.

[0184] While the KF analysis is useful, the is nonlinear when the noise increases. Also, if the tracking is a bit off (LUT and delay) the residual hopping signal in the loop may be large. Weneed an analysis that can 1) determine the PDF of the phase error in the loop 2) determine the mean time to cycle slip. For this we use the discrete phase state model. Assume M phase states from -180 deg to 180 deg and after each loop time constant of 1 μsec determine the transition probabilities. From these the transition matrix is set up and it is possible to determine the eigen decomposition which gives the probability distribution of the phase. From this we can determine the effective signal loss of the message signal demodulation. For a steady state PDF that is meaningful we need reflective states. Then by making the end states absorbing, we can determine the mean time to lose lock. If this mean time is significantly larger than the receive epoch, then the loss in m(t) demodulation will be negligible. By finding the threshold of SNR when this is not the case, then the required SNR is determined, and we can do a quantitative comparison of different schemes.

[0185] Note a limitation in this analysis is that it is first order Markov. A second order Markov may be represented which will be more accurate in terms of dynamics of the loop however, it is more effort to get right. We start with first order and move to second as needed.

[0186] T o set this up we have an error discriminator curve as shown in Figure G. One cycle is considered as moving outside of this range corresponds to a cycle slip. A cycle slip event is essentially a loss of the data frame. Hence, we do not have to consider multiple cycle slips, only the mean time to the first slip. We model the noise in the loop as normal that is independent from one 1 μsec time interval to the next. The analysis determines the spread of the PDF with time. A Monte Carlo simulation may also be done but having the time evolution of the PDF is much more insightful.

[0187] Then depending on the noise variance and level of e, we can determine transition probabilities. Typical discriminator looks like FIG. 35 where the ‘0’ in phase typically corresponds to 90 deg.

[0188] The normal noise variance (due to jamming interference and thermal) is the outcome of one-time constant that is integrated by the PLL loop filter and VCO. Based on the states we can create a matrix of transition probabilities. Then the model, seen in FIG. 36 is as follows. Top (1) is used to determine the mean time to cycle slip (iteration at which the sum of the probabilities of the end states exceeds 0.5). The bottom (0) uses reflective end states and is useful for generating the phase error PDF. Having this PDF gives the degradation of the SNR of m(t).

[0189] VCO Frequency Tracking

[0190] There will now be described an example of a transceiver for the that communicates between the RV platform and a GCS that uses a free running VCO with an analog input (albeit controlled from a digital DAC). Without the requirement to frequency or phase lock, the free running VCO may be FM modulated, an analog process, arbitrarily limited only by the bandwidth of the VCO frequency bias control (which is generally coupled to a varactor). Hence fast hopping is possible with the phase of the VCO remaining continuous, which gives the option of a matched filter receiver which is superior to an envelope magnitude receiver (AM) that ignores phase.

[0191] The challenge is then to have a means of tracking the remote vehicle VCO at the GCS. Let VCOdand VCOgdenote the remote vehicle and GCS VCO respectively. Then we need to have VCOgaccurately track the VCOdthrough the frequency hopping. For this we need tight time synchronization between the remote vehicle and the GCS. This allows the h0(t) hopping waveform to be synchronized. Define t as the remote vehicle time and t’ as the GCS time and then gt(t’) = t as the synchronizing function (i.e., clock and timing recovery). This involves tracking the propagation delay between the remote vehicle and GCS denoted as Dt.

[0192] Beyond the time synchronization, the VCOdwill drift relative to VCOgdue to temperature, aging, frequency changes do the power supply variations. As well there are unmodelled variation in the bias circuit. Afterall, the drive circuit is modulating a varactor voltage, a nonlinear device with memory. Additionally, we cannot assume that the VCOdhas had much of a bum in time and will age quickly during the first hour of operation. Burn-in time may be arranged at the factory but is an additional expense.

[0193] Implementation of Time Division Duplex (TDD) Links

[0194] A block diagram of an example of an RV circuit is shown in FIG. 37. The main feature of this example is that the VCO is free running for the uplink and downlink transmission of the TDD protocol. In uplink it is FM modulated with h0(t)+m(t) where h0(t) is the continuous hopping waveform that is used for both up and down link. m(t) represents the message signal with the data. For downlink, the VCO is modulated only with h0(t). The input signal is frequency locked to this at the GCS end such that the difference frequency is the IF frequency (in addition to themessage modulation). The IF contains a standard FM demodulation consisting of a pair of offset bandpass filters with a difference output which is really a bandpass differentiation circuit as indicated.

[0195] FIG. 37 shows a block diagram of an uplink transmission with VCO 20 used with an analog input from VCO 20 as it may be fast modulated and maintain coherence as it is modulated. H0(t) is the desired hopping, h(t) is the actual. GCS SDR maintains model of time difference and difference in h.

[0196] FIG. 38 shows a block diagram of an example downlink transmission from a model of RV VCO maintained in GSC it generates FM hopping signal based on h(t)+fIF+m(t). Hence RV VCO will accurately hit the narrowband IF frequency.

[0197] FIG. 39 shows the TDD (time division duplex) structure for uplinks and downlinks for a signal transmission between SA and SB.

[0198] A feature of the TDD continuous frequency hopping is that the hopping waveform of h0(t) may be continuous. Likewise, the m(t) modulation of the message may have no discontinuities. This modulates the VCO. Therefore, the signal remains phase coherent which is key to the matched filter operation and that h0(t) and m(t) may be independent. That is, we may hop slowly and have a fast m(t) or the other way round. There is no restriction.

[0199] Note that the RV VCO is being modelled by the GCS for most of the time, when modelling is interrupted by the downlink epoch. However, the coherence time of the VCO is much longer than this period or the latency of the GCS to update the model. Hence, we will be able to track and compensate for any frequency deviation error of the RV VCO as it evolves with time.

[0200] Selection of FM Modulation

[0201] The FM mode is considered in FIG. 40 for the modeAB. The SOM of system A is modulated by h0(t)+m(t) where m(t) represents a message signal that is initially unknown to system B. System B jointly estimates m(t) in addition to the SOM model of System A and the time difference.

[0202] Likewise in modeBA, systemB uses a message signal of m(t) to send back to systemA which is now in QER. A key point is that this transmission from systemB is h(t)+m(t) in which the larger h(t) component is coherent with the QER matched filter at the present time. Hence the only requirement is that m(t) is correctly demodulated / decoded.

[0203] VCO Frequency Tracking

[0204] As described above, the transceiver that is used in communicating between the RV platform and a GCS may use a free running VCO of the RV with an analog input (albeit controlled from a digital DAC).

[0205] VCO Pairing and Device Response Modelling

[0206] During the uplink epoch, h+m is the signal synthesized by the SA processing that is the tuning input into the free running VCO of SA with the actual frequency being uA(h+m) where uA() represents the frequency tuning function of the VCO which drifts over time. The LO frequency generated by SB may be described as uB(h)+fifwhere fifis generated by a frequency translation in the SB. Here uB() is the VCO tuning function of the SB. The VCO of SA and SB may be the same circuit architecture with equivalently valued components but they will not be precisely the same and therefore uA() ≠ uB().

[0207] Furthermore, uA() may drift in time differently than uB() primarily as the ambient factors and aging are different. In such a case, uA() must be paired or matched with uB() by some adaptive process. Assuming that the pre-launch calibration and configuration (PLC) has been done such that initially this pairing is accurate, the method of doing this is to have a ‘voltage warping function’ denoted by w() that satisfies the frequency matching relationship of uA(h(t)) =In PLC the time may be precisely synchronized, and h(t) is shared directly between SA and SB.

[0208] Assuming uA() and uB() are monotonic functions, determining the initial warping function of w() is straightforward.

[0209] For the adaptive tracking we have the frequency ofuB(w(h(t'))) + fif− uA(h(t) + m(t))This passes through the IF filter which is in two stages of a 10 MHz followed by a 1 MHz and then goes into a PLL. The bandwidth of the FM modulated m(t) will be about 1 MHz. We need to remove as much of the interference as possible before the FM demodulating PLL.

[0210] With respect to time synchronization, in the SA, the demodulation of m will result in a quality factor measure such as RMS noise in the demodulation eye at the sampling point. There is no additional processing or hardware required other than this quality value is sent to SB as part of the uplink data frame. SB processing estimates the delay time td and applies g(t’)-tdfor upstream and then g(t’)+td for downstream transmission. Additionally, a small dither may be added to the estimate of td which is somewhat equivalent to early / late gate tracking.

[0211] What Tracking is Possible

[0212] In practical applications, the high frequency slew rate of uA(h) may be an issue. Consider a linear ramp change in frequency of Af Hz that occurs in a time of At sec such that the frequency slew rate is AfAt Hz / sec. Now the tolerance of the frequency error is given as ATfand the tolerance of the timing error is defined as ATtthen the relation is.TtΔf / Δt= TforTt= TfΔt / ΔfTherefore if we have Afof 10 MHz and At of 1 μsec which implies that that the band hops 10x the IF bandwidth during the symbol time of m(t), and a tolerance of Af=100 kHz which is 10 percent of the IF bandwidth then we have Tt= 10 nsec. Is it reasonable to consider a timing that accurate? Base this answer on the motion of the RV which is fairly sluggish and that the controls to the RV are known. Also there are other aids that may be provided such as CV with optical flow of a downward facing camera as well as the IMU. There could also be other signals that may aid the position estimate of SA relative to SB. Next consider a rudimentary GPS receiver that may determine the position within several feet in a LOS position relative to the satellites. GPS has a lower SNR (thermal only as we are assuming hopping may get a clear signal for aportion of the time) and with that readily achieves a 1 foot error for a smoothing moving platform. Hence the satellite pseudo range must be determined with an accuracy of about 1 foot or the timing error within 1 nsec. Now there are pluses and minuses in this calculation. More GPS satellites are available however it is necessary to determine {x,y,z,t} and not just t as in our case. Furthermore, the sensitivity of timing errors in the uplink signal is much higher than that of the GPS CA signal. Therefore, the Fisher Information (Fl) of the uplink is much higher than that of the individual satellite-receiver link of GPS. Consequently, it is reasonable to assume that if the SNR of an uplink timing error based on a 100 msec epoch is no worse than the equivalent GPS link sampled over an equivalent epoch of 100 msec, that a 1 nsec timing error is reasonable to assume. This means that the Af of 100 MHz is reasonable in 1 μsec. Hence there will be some correlation of the hops covering 1 GHz span as this cannot be done in a 1 μsec epoch. Consider a frequency error ffedue to a timing error for some level of frequency hopping. Estimating fewill allow alignment of the h(t)-h(t’). The CRLB will give a lower bound to the estimate of the variance of determining fe.

[0213] It can be shown that an estimation variance of error frequency in MHz (as the sample rate is 1 MHz) is given asvar(f̂e) ≥(2π)2εsmpN3where esmpis the SNR per sample which is say about 1. This is pessimistic as the SNR required for the data decoding is several dB. Let’s start with N=100 which is 0.1 msec epoch which is a reasonable pilot header for the uplink frame. This gives a deviation of 559 Hz. Hence with a rate of megahertz slew per microsecond we can have sub-nanosecond delay estimation.

[0214] Next consider the warping problem. Assume that the PLC determines the initial t-t and w() such thatuB+ fif- uA(h(t)) = fifHowever, over time the uB(w( / i(t)) and uA(h(t)) will start drifting apart. Let the difference frequency be modelled asuB(w(h(t))) − uA(h(t)) = d0+d1h(t)If the parameters of d0and d1may be estimated by the GCS, then we will have a sense of how to correct the warping function w(h). But we do not measure frequency directly. Rather we havex(t) = Acos(θ + 2πfift + 2π∫0tt'(uB(w(h(t'))) − uA(h(t')))dt')We need to model this with the samplesy(t) = Âcos(θ̂ + 2πfift + 2π∫0tt'(d̂0+d̂1h(t))dt')Jowhere the parameters to be estimated are {A, θ, d0, d1} where {A, θ} are nuisance parameters. In practical processing terms, x(t) would be samples say at a 10 MHz rate by the SDR and then using correlation we will fit parameters to this. This is a numerical MLE.

[0215] It may be assumed that the frequency error is small enough that it may fit into the 1 MHz IF filter. For the example ignore the fif. As the number of samples is large, then the MLE becomes the asymptotic MLE implying that the variance of the estimators is given by the FIM.

[0216] The processing is to determine the MLE based on correlation which is really a least squares (LS) method, such as Python least squares. Shown in FIG. 41 is the result of fitted data model for noisy data.

[0217] After the difference curve is fitted, we can determine the correction to the w() as a function of h. A possible SB implementation is that we sample the uplink frame and then determine the frequency deviation as illustrated in FIG. 42. The uplink frame can have 1 msec of initial pilot. The amplitude is easily determined separately which leaves the B, C, D that may be estimated from the LS update. From this quadratic, we would get some typical relation as the following.

[0218] For regions r1and r3we would decrease w() slightly and for r2increase w(). For the tolerance gaps in-between, there would be no change.

[0219] What is left to consider is the feedback from SA regarding the timing. SB will adjust h(t+td) such that it is close as td of the uplink and downlink must be the same. We can do the same with a short pilot section followed by the payload and then an adjustment to this.

[0220] The strategy is to align t’ of SB to t of SA. Then fit td such that in the uplink correlate with h(t-td) and in the downlink transmit h(t+td). As we get feedback from SA in the downlink transmission regarding t+td and from SB regarding t-td we can determine tdand the timing errorof t-t’ simultaneously. That is, we have two conditionst − td= t' + tc− tdet = t' + tc+ tde− tdthe first for uplink and the second for downlink. Suppose that tdincreases then we must increase tde- tcby x. We send downlink and we must increase tde+ tcby y. So, we have[x; y] = [1 −1; 1 1][Δtde; Δtc]which clearly resolves the corrections. After these corrections we have the alignment of t’ to t and the estimate of td. But this does require that the measurement is done in both SA and SB.

[0221] The typical Allen Deviation for a TCXO is about 10’11for a T of 0.1 second as shown in FIG. 43. For the combination of the SA and SB TCXOs the variance would be more like 10’20for a T of 0.1 second. This increases linearly with time such that for 1000 seconds (which is a typical flight duration) it would be a variance of 10’16or a deviation of 10’8. This implies that unchecked the deviation would be 10 sec after 1000 seconds which is a total loss of synchronization as the time constant of h(t) is 1 μsec. For one second though it is 10’10which is 0.1 nsec per second. This is correctable as there are at least 10 frames per second. We can average the tccorrection over 10 second time constant which is then equivalent to 1 nsec. Hence in SA for the downlink we have 100 samples over which we can determine the tccorrection.

[0222] Modeling the instability of a TCXO in the time domain may be approximated with a simple random walk model. TCXOs are designed to reduce temperature-induced drift, but they still exhibit some degree of slow, cumulative frequency instability over time. This may be reasonably modeled as a random walk, where each step corresponds to a small variation in frequency or phase over a given interval. In the time domain, phase noise and frequency instability often result in small, cumulative timing errors that may be modeled as random perturbations added at each step. The random walk model corresponds to integrating white noise, which is often a good approximation for the random low-frequency fluctuations that dominate clock instability over longer intervals.

[0223] Next consider the variation of the actual delay time td. We will have a model of this based on sensors. Typically, a barometer is part of the I MU that provides an estimate of the relative altitude, and a camera with optical flow processing that provides an estimate of theground velocity. A magnetometer indicates direction. As such the radial velocity and accumulated radial distance may be estimated. However, there are complications. Accurate roll and pitch are necessary for the optical flow to provide a valid estimate. However, based on the flight controller I MU measurements and the flight control commands with some windspeed estimation it is possible to estimate td. It is a steady flight and not a high dynamic racing application. But td will change up to 50 nsec / sec which is far more variable than the TCXO. Hence, we must have sufficient information in one second to accurately determine td.

[0224] The process will be to assume tcis accurate such that td may be tracked from the pilot correlation. Application of a Kalman filter (KF) is appropriate as there is sensor data: changes in accelerometer, inertia, barometer, and power consumption of motors to name a few.

[0225] Kalman Filtering (KF)

[0226] Consider a simple 1 D example of an RV moving along x with a steady drift velocity but added is a Gaussian variation in each sampling interval. The update is white Gaussian with u ~ N(b, σu2) where b represents the drift and the variance of the random component motion.

[0227] At the nth interval there is a measurement of the absolute RV radial position of zn~ N(xn, σn2). Into the nth interval is the outcome of the previous state estimation of x̂n-1~ N(ze,n-1, σe,n-12). The five update steps are given as follows:Step 1: Prediction of xnprior to measurement based on ze,n-1and the bias of the update processxnp= ze,n-1+ bnStep 2: Variance of xM1 lpn= Ou'e2,n-dl +TOu’u2Step 3: Kalman gainK = (σ2e,n-1+ σ2u)(1 / (σ2e,n-1+ σ2u+ σ2n)) = Mpn / (Mpn+ σ2n)Step 4: State estimation correction with measurementxnc= xnp+ K(zn− xnp)Step 5: Posteriori varianceMnc= (1 − K)MnpFor the next iteration we then have x̂n~ N(ze,n, σ2e,n) = N(xnc, Mnc)

[0228] We use Matlab to simulate this KF. For this example, select b=0, σu=.02 and σn= 0.1.

[0229] Consider a particle with a drift velocity of vO but a gaussian random variation of v1 as measured in a sampling interval of T seconds. At each interval there is a noisy position measurement. Write a python / NumPy simulation of a Kalman Filter (KF) that estimates the position of the particle after each sampling interval.

[0230] With a Python script using NumPy to simulate a KF for estimating the position of a particle. The particle moves with a drift velocity (v0v_0v0) and a Gaussian random variation (v1v_1v1) is added to its velocity. A noisy measurement of the position is available at each sampling interval (TTT). The result is seen in FIG. 44.

[0001] Kalman Filter (KF) for tcand td

[0231] Suppose we sample every frame which implies every 0.1 seconds. For that interval range, the random walk model of the TCXO is valid as the Allen deviation increases at a rate of c√τ. The deviation for a typical TCXO will be on the order of 10’10for T=0.1 sec. The interval uncertainty is independent based on the deviation increasing at a constant rate. Hence the epoch of a 0.1 sec interval measured by the TCXO is then 10 psec. The timing uncertainty added is then 10’22sec2.

[0232] Suppose we have an overall interval of 1000 seconds which is 104intervals of 0.1 seconds each. Then the overall timing uncertainty is 10’18sec2which is a standard deviation of 1 nsec. The assumption of independent intervals is consistent with the constant deviation slope for larger intervals.

[0233] Fora typical TCXO, Temperature is random 1°C, deviation is about 0.5ppm for 100°Cand 5x1 O’9per °C. Large power supply changes of 5% is about 0.1 ppm change in frequency. So a change in LDO has a change in output voltage of about 5% over 100 degree temperature range, so.05% per °C. The TCXO frequency change will be about 10’10for a 1° change. So about 10’11change in a 0.1 sec interval, which is on the order of the Allen Deviation.

[0234] With various contributions to deviation, such as aging, base resonator, temperature, power supply, it is reasonable to consider 10’10Allen deviation for a 0.1 sec interval.

[0235] What is the SB uncertainty in measurement of the pilot block of duration of 0.1 msec? This is 100 time constants of frequency hopping. Using CRLB analysis, we sample at 100 MHz, so it is reasonable to assume 10 nsec error. CRLB may be negligible as the frequency variation over this interval is large, so it either fits or does not fit. Question is more what is the probability of a missed correlation? Ratio of the energy in the 0.1msec signal epoch to noise. We have a measure of the SNR and hence the probability of error.

[0236] Another input uncertainty is td. Say the RV SA moves at 50 nsec / sec (34 mph). Change of 5 nsec every interval. Now we can assume that the SA takes 1 second to stop so it takes 10 intervals to change rate to zero. Uncertainty is about 0.5 nsec. That is, if the RV is going at a 5 nsec change rate in one interval, the next interval may be 4.5 nsec / sec rate change.

[0237] For the mbol period of 1 μsec we would be misaligned in tuning with change in temperature and supply voltage as seen in FIG. 45.

[0238] Consider a requirement of Start with something modest: 1 MHz in 1 μsec. This means that we have a new hopping frequency outside the 1 MHz band every bit sample. This is a 1012Hz / sec slew rate. Now we may be 105 Hz off and still make it through the IF filter. This implies that the timing error may be up to 100 nsec which is certainly possible. In GPS receivers we readily track timing down to 1 nsec which means we may have 100 MHz change in 1 μsec.

[0239] 1 nsec is doable if the radial velocity may be tracked. Note that we have the controls known, all wind speeds etc. Hence the radial distance to the RV may be estimated to within 1 foot = 1 nsec.

[0240] In this section we will quantify what is possible with the run time tracking. It will be assumed that the PLC perfectly synchronizes the clock such that t=g(t’) where g() is somefunction accommodating initial offset of t-t’ and the drift of t-t’. The frequency output of the IF of SB was given as followsuB(w(h(g(t')))) + fif- uA(h(t - td) + m(t - td))From this we need to determine an update of the warping model w(), the timing error oft - td- t'and m(t).

[0241] There are two parts to this problem. The first problem is the determination of what is in theory possible to track, for which we may use an aspect of estimation theory. The second problem is given what is in theory possible, to design an implementable processing system with the GCS that will keep track of the changes in the tuning curves and time drift.

[0242] For the first problem we will assume initial perfect knowledge of w() and m(t) and that g(t’)=t. Then determine the sensitivity of variation of parameters that need to be estimated to track w(). Then consider the noise in the samples followed by application of the FIM.

[0243] The assessment of what is possible is based on the CRLB analysis. Consider first the simplified problem of uA(h) drifting relative to uB(w(h)) with constant h. We assume a sampling say during the uplink epoch of duration of 10 msec. The FIM is given asand the variance of the parameter estimation is1var(T)This means that there exists an estimator that may in principle be part of the GCS processing which has a minimum variance given by var(T).

[0244] Consider the Fisher information of a Gaussian variable with the parameter being the bias A x ~ N(x|A, σ²)A, σ2J) where 0 = A. Hence x ~1 / √(2πσ2) e-½(1 / σ²)(x-A)²and√2πσ2dy / dAd / dA ln(f(x|A)) =such thatJA= E[((x-A) / σ2)2] = (1 / σ4)E[(x-A)2] = 1 / σ2JA= E[((x-A) / σ2)2] = (1 / σ4)E[(x-A)2] = 1 / σ2such that the CRB isvar(T(x)) ≥ JA-1= σ2

[0245] PLC Generation of w()

[0246] Consider the PLC first. Initially m(t)=0 and h(t) is static. We use the 10 MHz detector to determine w(h(t)) for each static frequency of the VCO. Then the 1 MHz detector is used for a refined value of w(). Next the w(t) is smoothed with an interpolating function which is stored in the LUT.

[0247] The hopping range is 1 GHz, and we need accuracy within a target 100kHz. Hence an accuracy of 100 ppm relative to the hopping range. Relative to the absolute frequency we must be 20 ppm. Note that the bandwidth of 1 MHz will not provide 100 kHz resolution the centering of the frequencyuB(w(h)) + fif2- uA(h)The initial TCXO may be trained to within 1 ppm. 1 ppm is 1.5 MHz offset at the start. However, this is offset by the VCO training of w().

[0248] Can be placed within 100kHz target band 4902 with dithering of w(h) across range 4904 that determines the band edges of the 1 MHz BPF and this will provide the initial VCO characterization to within 10 MHz. Then the 1 MHz detector is used as a refined search. This results in the setting the center tracking frequency shown in FIG. 46.

[0249] Run Time Adaptive Calibration

[0250] We use a typical specification of 15 MHz / volt for the power supply. We want the PS modulation to less than 1 kHz which is then 1 part in 1,5e4. Hence the power supply spectral noise should be below 80 dB. The noise will primarily be in the switching supply frequency. Also, the VCO power will be a function of the frequency. Hence at a 1 MHz rate the power output will be modulated. But this may be reduced with an active LDO and plenty of capacitance. But the ripple may be no more than 1 mV.

[0251] VCO input tuning capacitance is 20 pF. At 1 MHz this is an impedance of 50 kohm. The opamp output current into the VCO, considering a 10Vpp swing will be on the order of 0.2mA.

[0252] We must be mindful of frequency pulling due to the load. Must have VCO, attenuator, gain block with good s12 for isolation. There is no temperature data, but this may be assumed to be 100 kHz / °C. Change in temperature may be assumed to be a time constant in the seconds range.

[0253] Can be placed within 100kHz with dithering of w(h) that determines the band edges of the 1 MHz BPF and this will provide the initial VCO characterization to within 10 MHz. Then the 1 MHz detector is used as a refined search.

[0254] Passing through the 10 MHz IF filter with a bandwidth that can accommodate the m(t) FM modulation but not large frequency errors in the difference frequency.

[0255] An initial approach will be to assume initially that we have a PLL doing the FM demodulation. Ideally the frequency isfif- gAuAm(t)Where gAis s slope of uA(h( )) which varies with h(t). In the design of the VCO, there will be an attempt to linearize this. Note that in the processing we may also generate an approximation to ggA(h) as a secondary correction. If the m(t) modulation is simple (such as bi-level encoding) then this is not necessary. If a multi-level encoding is used, then correction may be necessary.

[0256] What drift will we have? Temperature and voltage which will tend to translate the entire tuning curve as seen in FIG. 45.

[0257] Next consider the PLL which has a loop bandwidth that is fast enough to track the difference in hopping ofuB(w(h(t'))) + fif2- uA(h(t - td))We quantize the output of the PLL with the objective of nullinguB(w(h)) - uA(h)which may be done as h is known.

[0258] A key point is that if it is a simple translation, then the frequency translation will be a simple approximate function offif + A / i(t) — Bm(t)Where A and B are positive constants. The PLL will track this, and from the demodulated output we can extract A and B based on decision feedback. m(t) would have some coding structure or pilot that makes this possible.

[0259] We can also use the PLL output to synchronize time as in h(t’)-h(t-td). We don’t have t or td, so all that we can do is to align t’ to t-td for the uplink communications.

[0260] Block Diagrams of RV

[0261] FIG. 47 shows a block diagram of an example of an RV in transmit mode, while FIG.48 shows a block diagram of an example of an RV in receive mode that receives input from QER of receiver and outputs demodulated m. As can be seen, VCO may be used both as the modulator for generating the transmit signal, and as the local oscillator to assist in demodulating the received signal. The RV may have a different demodulator design. The processor may be used to control the VCO to generate the appropriate signals, and may be used to interpret the message signal either being received or transmitted. While a single processor 412 is shown throughout this disclosure, it will be understood that this is representative only, and that different processors may be used for different purposes. Each function described herein may be performed by a computing module, where a single module may perform more than one function, or a function may be divided between different modules.

[0262] In this patent document, the word "comprising" is used in its non-limiting sense to mean that items following the word are included, but items not specifically mentioned are not excluded. A reference to an element by the indefinite article "a" does not exclude the possibility that more than one of the elements is present, unless the context clearly requires that there be one and only one of the elements.

[0263] The scope of the following claims should not be limited by the preferred embodiments set forth in the examples above and in the drawings, but should be given the broadestinterpretation consistent with the description as a whole.

Claims

What is Claimed is:

1. A radio architecture, comprising:a first transceiver in wireless communication with a second transceiver, wherein: the first transceiver comprises:a first antenna;a first signal processing module connected to the antenna;a free running oscillator connected to the antenna and the first signal processing module, the free running oscillator being frequency tunable; anda first control module connected to the free running oscillator, the first control module comprising instructions to tune the free running oscillator such that:in a transmit mode, the free running oscillator generates a transmit signal that comprises a message signal and a hopping signal, wherein the transmit signal is continuously smooth and the hopping signal is pseudorandom; andin a receive mode, the free running oscillator generates the hopping signal for use by the first signal processing module to extract a received message signal from a receive signal; andthe second transceiver comprises:a second antenna;a second signal processing module connected to the antenna;a tunable oscillator connected to the second signal processing module; and a second control module that comprises instructions to frequency-match the tunable oscillator to the free running oscillator using the hopping signal.

2. The radio architecture of claim 1, wherein the free running oscillator comprises a voltage controlled oscillator or a Q-enhanced resonator.

3. The radio architecture of claim 1, wherein the first transceiver is an airborne transceiver, and the second transceiver is a ground-based transceiver.

4. The radio architecture of claim 1, wherein the first transceiver is a mobile transceiver and the second transceiver is a base transceiver.

5. The radio architecture of claim 1, wherein the second transceiver comprises a second free running oscillator.

6. The radio architecture of claim 1, wherein the first control module and the second control module are programmed with instructions to cause the first transceiver and the second transceiver to communicate using time division duplex.

7. The radio architecture of claim 1, wherein the hopping signal comprises a central frequency that changes within a hopping bandwidth.

8. The radio architecture of claim 1, wherein the first signal processing module comprises a mixer and, in the receive mode, the free running oscillator is adapted to act as a local oscillator for the mixer.

9. The radio architecture of claim 1, wherein, in the second control module, the instructions to frequency-match the tunable oscillator further comprise calculating a warping function that represents a difference between an expected frequency and an actual frequency of the free running oscillator.

10. The radio architecture of claim 9, wherein the warping function is adapted to correct for at least a frequency drift of the free running oscillator and a transmission time delay between the first transceiver and the second transceiver.

11. The radio architecture of claim 9, wherein the warping function is adjusted based on a calibration signal transmitted between the transmit signal and the receive signal, the calibration signal comprising the hopping signal in isolation.

12. The radio architecture of claim 1, wherein the first signal processing module comprises at least one signal filter and a demodulator.

13. The radio architecture of claim 1, wherein the first signal processing module comprises a phased locked loop (PLL).

14. A method of wireless communication, comprising the steps of:providing a first transceiver that comprises:a first antenna;a first signal processing module connected to the antenna;a free running oscillator connected to the antenna and the first signal processing module, the free running oscillator being frequency tunable; anda first control module connected to tune the free running oscillator; providing a second transceiver that comprises:a second antenna; anda signal processing module connected to the antenna, the signal processing module comprising a tunable oscillator;in the first transceiver, tuning the free running oscillator such that, in a transmit mode, the free running oscillator generates a transmit signal that comprises a message signal and a hopping signal, the hopping signal comprising a continuously smooth pseudorandom signal, and in a receive mode, the free running oscillator downconverts a receive signal received via the antenna using the hopping signal; andin the second transceiver, tuning the tunable oscillator to frequency-match the tunable oscillator to the free running oscillator.

15. The method of claim 14, wherein the free running oscillator comprises a voltage controlled oscillator or a Q-enhanced resonator.

16. The radio architecture of claim 1, wherein the first transceiver is an airborne transceiver, and the second transceiver is a ground-based transceiver.

17. The radio architecture of claim 1, wherein the first transceiver is a mobile transceiver and the second transceiver is a base transceiver.

18. The method of claim 14, wherein the second transceiver comprises a second free running oscillator.

19. The method of claim 14, further comprising the step of using time division duplex to communicate between the first transceiver and the second transceiver.

20. The method of claim 14, wherein the tunable oscillator is tuned based on the hopping signal, the hopping signal comprising a central frequency that changes within a hopping bandwidth.

21. The method of claim 14, wherein the first transceiver comprises a mixer and, in the receive mode, the free running oscillator acts as a local oscillator.

22. The method of claim 14, wherein, in the second control module, frequency-matching the tunable oscillator comprise the use of a warping function that represents a difference between an intended frequency and an actual frequency of the free running oscillator.

23. The method of claim 22, further comprising the step of updating the warping function based on a received signal from the first transceiver.

24. The method of claim 22, wherein the warping function is stored in a look up table.

25. The method of claim 22, wherein the warping function is adapted to correct for frequency drift of the free running oscillator and transmission time delay.

26. The method of claim 22, wherein the warping function is adjusted based on a calibration signal transmitted between the transmit signal and the receive signal.

27. The method of claim 14, wherein, in the transmit mode, the first control module maintains a coherent phase as the free running oscillator is tuned between frequencies.

28. The method of claim 14, wherein the first signal processing module comprises at least one signal filter and a demodulator.

29. The method of claim 28, wherein the demodulator further comprises a phased locked loop (PLL).

30. A radio architecture, comprising:a first transceiver in communication with a second transceiver, wherein:the first transceiver comprises:a first signal processing module connected to receive a receive signal;a free running oscillator connected to the first signal processing module, the free running oscillator being frequency tunable; anda first control module connected to the free running oscillator, the first control module comprising instructions to tune the free running oscillator such that:in a transmit mode, the free running oscillator generates a transmitsignal to the second transceiver that comprises a message signal and a hopping signal, wherein the transmit signal is continuously smooth and the hopping signal is pseudorandom; andin a receive mode, the free running oscillator generates the hopping signal for use by the first signal processing module to extract a received message signal from the receive signal; andthe second transceiver comprises:a second signal processing module connected to receive the transmit signal from the first transceiver,a tunable oscillator connected to the second signal processing module; and a second control module that comprises instructions to frequency-match the tunable oscillator to the free running oscillator using the hopping signal.

Images