An overview of the effects of out-of-band interference on GNSS receivers

This paper addresses the wide range of mechanisms through which out-of-band interference can disrupt the functioning of GNSS receivers. These mechanisms include saturation and desensitization of front-end low noise amplifiers, mixers and other circuitry; reciprocal mixing effects that arise from the fact that receivers cannot generate a perfect tone to down-convert the desired signals; intermodulation products; aliasing of out-of-band emissions that remain after filtering into the receiver's passband; and the reception of in-band (to GNSS) emissions that are always present due to imperfections in the signal generation and filtering of the interfering system. These mechanisms are described in detail and mitigation approaches for each are discussed.

Figure 1 depicts a representative radio-frequency (RF) front-end design for a modern digital GNSS receiver.This design, which is based upon a typical configuration for an aviation receiver, will be used to illustrate the various effects of out-of-band interference that are discussed throughout this paper.
The active antenna unit (top portion of Figure 1) includes a passive patch antenna element tuned to the operating frequency, limiter (to provide lightning protection), two dielectric resonator (ceramic) filters, and a low noise amplifier (LNA).The active antenna unit is connected to the receiver via a length of cable.The receiver unit itself includes an input filter, comparable to those in the antenna, a limiter, LNA, a pre-mixer filter, followed by a mixer to down-convert the received signal using a local oscillator (LO) to some convenient intermediate frequency (IF).Following down-conversion, the IF signal is filtered by a surface acoustic wave (SAW) filter and amplified for levels suitable to feed an analog-to-digital converter (A/D).The A/D digitizes the signal for further signal processing.Automatic gain control (AGC) ensures a proper A/D input signal level.
It is important to note that the design in Figure 1 is only illustrative.Across the entire set of fielded GNSS receivers, the designs vary greatly (see, eg, Van Dierendonck 3 ).Some configurations use passive antennas (ie, the external antenna unit only contains the passive antenna element in a protective casing, or radome, with a connector).The amount of filtering within active antennas may vary tremendously from product to product, and the amount of filtering and filter technologies used within the receiver unit may also vary tremendously.Some receivers may use two or three stages of down-conversion versus the single-stage illustrated.Many receivers sample or subsample a single IF signal as shown in Figure 1 using A/D techniques, whereas others use a pair of mixers in the final analog down-conversion to bring inphase and quadraphase components of the received signal fully to baseband before separately sampling each of these two paths.Lastly, Figure 1 only depicts the front-end for one center frequency.A receiver processing GNSS signals upon multiple frequencies would require additional front-end hardware to implement parallel signal paths.

| Components
Despite the wide variation, noted above, in GNSS receiver designs, the basic components depicted in Figure 1 are found in virtually all receivers.The behavior of these components influences the overall performance of the receiver in the presence of strong out-of-band interference.

| Passive antenna element
Many GNSS antenna designs (eg, microstrip patch or stacked-patch) are resonant at one or more desired frequencies, whereas others (eg, conical spiral) are broadband.Resonant antennas can provide substantial attenuation to far out-of-band signals.The gain pattern and polarization characteristics of the antenna are also important parameters since they influence the received power levels for both the desired signals and interference.It should be noted that performance characteristics of an antenna may be significantly different when the antenna is installed and operated as compared with when it is measured in a test chamber by the manufacturer.

| Limiter
Limiters are implemented in most high-quality active antennas and receivers.A typical design uses a shunt PIN diode to protect subsequent components against high input power levels, lightning, and electrostatic discharge events.(PIN refers to a semiconductor with p-type and n-type outer layers and an intrinsic layer in-between.)Depending on the applied RF power level, the device appears as an open circuit at low power levels, somewhat resistive at moderate power levels as the diode enters forward bias conditions, and a very low impedance at high power levels that cause high forward bias.While inherently nonlinear, the levels required to enter the nonlinear region are normally far higher than those occurring during interference events.

| RF and IF filters
A wide variety of RF and IF filters are used across the enormous number of fielded GNSS receivers.The most commonly used filter technologies are described in detail in the later section of this paper on mitigation approaches.
RF filters isolate the input spectrum to the desired passband, rejecting noise and signals from outside the passband.Important system level characteristics include rejection at known interference signal frequencies and rejection at image frequencies of the receiver for both interference signals and thermal noise.
IF filters isolate the channel spectrum to a much tighter bandwidth than the RF filters are capable of, thus ensuring close-in interference sources are rejected, as well as minimizing the noise bandwidth of the analog signal processing stages prior to the A/D.
Often, the most susceptible region across the bandwidth of a receiver is that region just outside the IF passband of the receiver.Here, while the IF filter would reject the signal, the RF filters have only weak attenuation so that even low level interference signals will pass virtually unattenuated through the RF and IF stages before the IF filter.Active stages just prior to the IF filter then risk being driven into saturation.

| LNA
Current LNA devices tend to use silicon-germanium (SiGe) heterojunction bipolar transistor (HBT) technology in either discrete or monolithic microwave integrated circuit (MMIC) packages.Gallium arsenide (GaAs) devices remain popular in both discrete and MMIC formats.When the LNA is part of a larger scale integrated receiver, SiGe, CMOS, or more recently combined SiGe-CMOS technologies are used.
Discrete devices have better noise figure performance, allow tailoring of other important performance parameters (gain, compression point, etc) to the application, and are often preferred for low-volume, high-precision receivers.However, there are a large number of MMIC LNA devices optimized for GNSS frequencies which offer size and complexity advantages over discrete devices.A single device has between 15 and 20 dB of gain at L band, so it is necessary to cascade two devices to obtain the LNA gain blocks described here.For this configuration, the first device would be chosen for optimal noise figure whereas the second would be chosen for compression characteristics.
As will be discussed later, LNAs operate in a nominally linear manner for small input voltages, but become increasingly nonlinear in the presence of increasingly strong interference.LNA nonlinearity is commonly characterized by specifications for the 1-dB compression point and third-order intercept point, which will be described later.Typical GNSS LNA output 1-dB compression points range from 0 to +10 dBm.The output third-order intercept point for a typical active device (LNA or mixer) is 10 to 15 dB higher than the output 1-dB compression point.

| Mixers
Virtually, all mixers used in current receiver designs are active mixer devices in either MMIC form, or within larger scale receiver integrated circuits (ICs).These devices use a variety of mixer configurations ranging from nonlinear amplification in single active GaAs or HBT transistors to Gilbert cell 4 analog multiplier configurations and are integrated with gain stages to overcome the inherent loss of the mixing process and to allow low level local oscillator (LO) signal levels to be used.
Basically, any device that has an input/output transfer function (V RF /V IF ) that can be modulated or multiplied by a voltage at a third terminal (V LO ) will contain a component V IF = K × V RF × V LO at its output terminals.(K will vary among devices.)Mixing results are illustrated below assuming sinusoidal waves for both the RF and LO signals: The IF output of the mixer contains components at the sum and difference of the LO and RF frequencies.For down-conversion, it is the signal at the difference frequency that is desired, and this signal is isolated at the output of the mixer using a low-pass or band-pass filter.
This relationship holds for numerous mixer implementations ranging from weak modulation of amplifier bias levels by an LO signal to complete high level on/off switching of the signal path at the LO rate.Using switches for the active device is efficient and simple where the implementation can be a single switch using a diode or transistor.More complex arrangements of devices, including pairs of switches driven in opposition with up to four devices in a balanced ring arrangement, can be used to provide advantages in cancelling unwanted spurious mixer products.The Gilbert cell 4 is a very commonly used integrated mixer configuration that implements a full four-quadrant analog multiplier.
From an interference perspective, active mixers typically have slightly lower 1-dB compression (P 1dB ) points, −10 dBm to 0 dBm referenced to their output, and accordingly lower third-order intermodulation (IP 3 ) behavior as amplifiers and are analyzed using the same techniques.

| Local oscillators
Phase locked loops (PLLs) are normally used to generate the local oscillator signals.In recent designs, the PLL implementations in integrated devices contain all the digital and analog components.Two flavors of PLLs exist: those with fractional frequency dividers and the more common implementations with integer dividers.Each generates a clean LO signal using a voltage-controlled oscillator (VCO), but some attention must be given to ensure that the reference frequency sidebands are adequately suppressed.For fractional-divider based PLLs, this can get quite complicated.
Generally speaking, frequency sidebands can occur in one of two ways.First, as shown in Figure 2, PLLs effectively generate their output signal by taking an external frequency source (eg, a crystal oscillator), dividing it down to some common reference frequency, usually in the tens of kHz region, and essentially multiplying it back up again to the necessary LO frequency.As the phase detector utilizes narrow impulses to control the VCO frequency, remnants of this reference frequency and its harmonics modulate the LO signal and show up as symmetrical sidebands at the PLL output.This frequency modulation is transferred to the received GNSS signal through the mixing process.However, this is not generally a concern because it occurs at very high frequencies relative to the baseband signal processing of GNSS information.It is simply a matter of good design practice to keep the level of these to the −50 or −60 dBc range.
The second type of LO sidebands occur due to internal leakage within the PLL devices or due to inadequate isolation around the PLL circuits.These spurious signals can also modulate the LO signal resulting in symmetrical sidebands.Again, good receiver design practices need to be employed to keep these spurious signals to reasonably low values, which are on the order of −60 to −80 dBc.
Figure 3 is a measured spectrum from a GNSS receiver local oscillator which shows both types of sidebands on a GNSS receiver LO signal.The reference frequency sidebands begin at 250 kHz and decrease with each integer related harmonic.Leakage sidebands can be seen at 2.5 MHz, about 3 MHz, and at the receiver reference clock frequency of 9.96 MHz.
Direct digital synthesizers (DDS) are becoming a common method to generate LO signals.DDS signals are generated by digital means and converted to an analog waveform prior to being output.They can achieve extremely fine frequency resolution but are rich in spurious signals, which can occur at virtually any frequency, albeit at low levels.This means that all effects including those affecting GNSS signal processing, receiver selfinterference, and reciprocal mixing of input interference must be considered when evaluating a DDS LO.Older style, simple receivers used one bit, hard limiting, which is not really AGC in that there is no feedback and control mechanism.The receive path signal is simply amplified until it clips and the clipped signal is then sampled as a single bit digital representation of the input signal and processed accordingly.This works, but the receive signals are degraded in the process.

| AGC amplifiers
Most modern GNSS receivers use multi-bit A/D devices to digitize the input signal.This requires linear amplification in the received signal path with the ability to center the level at the A/D to the required receiver setpoint.In the analog domain, an AGC amplifier, or more properly, a variable gain amplifier (VGA) is used for this purpose.Control of the VGA is by an AGC loop, with feedback coming from the GNSS signal processing blocks.
VGAs can be implemented by controlling the bias of a simple transistor amplifier or by more complex implementations that contain several stages of amplifiers and complex networks to idealize the gain control curve.Figure 4 shows examples of receiver gain measurements for two receivers.The first receiver uses simple transistor bias control and a second a dedicated linear-in-dB RF integrated circuit (RFIC) device.
While the overall gain control range is similar, the simple transistor bias control has a far less ideal gain control curve.Worse yet, in the context of interference performance is the behavior of the compression point of the VGA stage.The compression point is a function of the bias level, being high for high bias current levels and low for low bias current levels.Strong interference levels at the input of the VGA (which would take control of the AGC loop if it were high enough) would cause the AGC loop to reduce the VGA gain by reducing the VGA bias.This is a situation of high interference levels and low compression points and is therefore problematic not only for interference that falls inside the IF band but also for out-of-band interference in receivers with IF filters that do not have sharp passband-to-stopband transitions or in receivers where the AGC amplifier precedes the IF filter.
RFIC devices strive to implement a far more ideal gain control curve than that of a simple transistor under bias gain control.However, depending on the particular implementation, they will also have other parameter variations as the gain control is varied.An idealized model of a VGA would contain a fixed amplifier plus variable attenuator.If the variable attenuator is in front of the amplifier, the VGA noise figure would be a function of the control voltage and the output compression point would remain constant.This is the preferred model for a GNSS receiver.However, some devices incorporate some gain prior to the variable attenuator making the composite device compression point a function of gain, which is the main weakness of the simple transistor circuit.

| Saturation and desensitization
Many GNSS receiver components may be accurately modeled using idealized models and linear system theory in the absence of strong interference.For instance, consider the simple LNA system model shown in Figure 5.The LNA takes an input voltage, x(t), which is typically the filtered output of a passive antenna element, and provides an output voltage, y(t), with a nominal power gain of G (in linear units).In the absence of interference, the LNA is well-modeled as ideal, ie, its input-output voltage characteristics may be accurately described as where a 1 = ffiffiffi ffi G p .For large input voltages, the LNA begins to saturate and is no longer well-modeled by Equation ( 2).A truncated Taylor series expansion is often used (see, eg, Pozar 5 ) as a more accurate model for the output voltage: For example, Figure 6 shows the voltage input-output characteristics of an LNA modeled using Equation ( 3) with N = 5 and the following coefficients: where R is the reference load resistance across which the voltage is measured, presumed to be 50 Ω.The a 2 and a 4 coefficients were set to zero, as is usual practice, assuming an input-output voltage function with odd symmetry.
The a 1 and a 3 coefficients were selected to achieve a 1-dB input compression point of −22.2 dBm, an output thirdorder intercept point of +20 dBm, and a nominal gain of 34.5 dB, which along with an assumption of 4.9 dB limiter and filter losses would result in an overall active antenna assembly compliant with relevant specifications for active airborne GNSS antennas from RTCA. 6 The a 5 coefficient was selected solely to achieve a "flat" output voltage for the maximum input voltage shown in the figure, as would be expected for a saturating amplifier.The input-output power characteristic of the modeled LNA is shown in Figure 7.The figure shows the 1-dB compression point (referenced to both input and output powers), where the output level of the LNA is 1-dB less than it would have been if the LNA maintained its nominal linear gain characteristic (ie, output power equals input power plus 34.5 dB) indefinitely as input power is increased.The output power at the 1-dB compression point for this particular LNA is +11.3 dBm, which is slightly higher than the earlier-mentioned 0 to +10 dBm level that is more typical for mass-market GNSS LNAs.
If the input voltage to the antenna LNA consists of just the desired GNSS signals and noise, saturation does not occur.Prior to digitization and correlation, the GNSS signals are buried by the noise, so it suffices to focus attention on only the noise levels to understand the overall signal levels seen throughout the front-end.For a typical noise density of −201.5 dBW/Hz referenced to the output of the passive antenna terminals, the noise power referenced to this point over the final IF bandwidth of 20 MHz bandwidth is −98.5 dBm. Figure 1 indicates the noise level for various points in the front-end.The LNA in the active antenna sees approximately −101 dBm at its input, and the output power of the entire active antenna assembly is −68.9 dBm.After a 10-dB cable loss and 2.7 dB losses for a filter and limiter, the first receiver LNA sees −81.6 dBm at its input.Both the mixer input and AGC VGA input are at −53.8 dBm, the final amplifier input is at −30 dBm.
In the presence of out-of-band interference with increasing strength, one or more of the active devices in the receiver modeled in Figure 1 will at some point begin to saturate.If the interference is far away in frequency from the desired band, saturation of the first LNA within the active antenna is most likely.The amplifiers and mixers further downstream will be provided some additional protection by the filtering stages that precede them, provided that the accumulated filter attenuation outpaces the accumulated gain as the signal moves along the front-end.For out-of-band interference that is closer to the desired band, the later stages are more likely to saturate first.
When saturation occurs, there are several deleterious effects on receiver operation that are together referred to as desensitization.First, the gain provided to the desired signal decreases.Importantly, in the presence of a strong out-of-band interference source (often referred to as a blocker), the gain provided by an active device to the desired, in-band signal is less than that provided to the interference source as may be observed by evaluating Equation (3) in the presence of the combination of a strong tone (interference) at one frequency and a weak tone (desired signal) at another. 7,8Second, as is well known, the output of the nonlinearity is distorted. 9For non-CW interference, the resulting interference power spectrum is generally spread over a wider range of frequencies, which often results in more energy entering into the receiver passband.The receiver simultaneously suffers a decrease in recovered power from the desired signal.Third, the overall noise figure of the receiver front-end degrades due to the diminishing desired signal gain, as well as due to an increase in the active device's individual noise figure.The latter factor occurs as a result of two mechanisms: (1) in saturation, the operating conditions of the device are no longer as intended to minimize noise figure, and (2) the nonlinear behavior results in the conversion of noise generated at other frequencies within the device to the desired band. 7,10,11

| Harmonics and intermodulation
The same nonlinearities in LNAs and active mixers that cause receiver desensitization can yield harmonics and intermodulation products due to the presence of strong out-of-band interference at their inputs. 5For instance, consider the behavior of an active component with the input-output characteristic of Equation (3) when the input is a sinusoid: A linear device, for which a i = 0 for all i > 1, would only output a sinusoid at the input frequency.For a nonlinear device, the higher-order terms in (5) give rise to sinusoids at integer multiples of the input frequencies, which are referred to as harmonics.Harmonics generated by nonlinear devices in GNSS receivers are not generally problematic since only interference signals that are somewhat close to the GNSS center frequency will survive front-end filtering, and the harmonics of such frequencies will be far outside the passband of downstream filtering.
When signals at two or more frequencies are present at the input of a nonlinear device, then intermodulation products result.For instance, with two tones at the input, the output of a nonlinear device that is well-modeled by Equation ( 3) is Using trigonometric identities, it may readily be demonstrated that the i = 2 output term includes energy at four frequencies referred to as the second-order intermodulation products: 2ω 1 , 2ω 2 , and ω 1 ± ω 2 , and the i = 3 term includes energy at six frequencies (third-order intermodulation products): 3ω 1 , 2ω 1 ± ω 2 , 2ω 2 ± ω 1 , and 3ω 2 .The i = 3 term is commonly the most troublesome within GNSS or other mass-market receivers, since the two frequencies 2ω 1 − ω 2 and 2ω 2 − ω 1 can fall within the receiver passband for strong interference at two frequencies that are sufficiently close to the desired signal center frequency to survive earlier band-pass filtering.Importantly, the third-order intermodulation products grow at three times the rate of growth of the level of the input tones, ie, for every 1-dB increase in the input, there is a 3-dB increase in the output third-order intermodulation product level.
If the interference of interest is a modulated signal and not just a tone, the analysis approach above must be modified accordingly.For the same reasons mentioned above, third-order intermodulation products are of the greatest concern and will be the focus of attention for the remaining discussion.Noting that the input-output characteristics are virtually unchanged for input power levels below −20 dBm by truncating the Taylor series to N = 3, and as noted earlier, the a 2 coefficient is normally very small for devices exhibiting odd-symmetry, and here, we focus on the simpler input-output voltage model: with the input voltage presumed to be well-modeled as a Gaussian, wide sense stationary random process.The output voltage autocorrelation may be determined as where R x (τ) is the input voltage autocorrelation and σ x 2 = R x (0) is the variance of x(t) (ie, input power aside from a possible scale factor for a non-unity resistance value).
From Equation ( 8), the following expression may be derived to relate the output power spectrum, S y (f), of the LNA to its input power spectrum, S x (f): where * is the convolution operator.The reader is cautioned that Equation ( 9) only applies for signals that are noise-like in the sense that they are well-modeled as Gaussian and wide-sense stationary.Modulated signals are most often cyclostationary and their cyclostationarity may result in a very different power spectrum at the output of a nonlinearity.For instance, bandlimited noise fed through a square law device will exhibit a power spectrum at the output that is the input power spectrum convolved with itself, as can be predicted from a derivation similar to that leading up to Equation (9).A binary phase shift keyed signal fed through the same square law device, however, will not have its power spectrum spread.The squaring in this case will result in the wipe-off of the data and a pure tone will result.
As an example of an interference problem due to third-order intermodulation products, consider the impact on the receiver model of Figure 1 in the presence of a pair of equal-power, 5-MHz bandwidth interfering signals centered at 1528.8 MHz and 1552.7 MHz.Using the attenuation characteristics in Figure 8 for the preselection filter (ie, the filter immediately after the passive antenna in Figure 1), Figure 9 shows the input and output power spectrum of the active antenna LNA if the two interference signals are seen at equal levels at the output of the passive antenna element with a total power of −45 dBm.(Note that we presume, for this illustration, that the passive antenna element equally attenuates each interfering signal, whereas in practice it would likely provide more attenuation to the lower frequency signal.)The preselection filter attenuates both interfering signals, but the lower-frequency signal that is further from the desired signal is attenuated more.At the LNA input, a total power level of −57.5 dBm is seen.This level is sufficiently low that the LNA does not experience a problem with saturation and the output power spectrum looks identical to the input spectrum except that it exhibits a 34.5 dB power increase due to the LNA gain.
If the interference is seen at a stronger level, the LNA will saturate and produce intermodulation products.This condition is shown in Figure 10.In this case, the interfering signals are received at a power level of −25 dBm at the output terminals of the passive antenna element.After the preselection filter and limiter, the LNA sees an input level of −37.6 dBm.A third-order intermodulation product is clearly seen centered at 2 × 1552.7-1528.8= 1576.6MHz, very close to the desired signal center frequency of 1575.42MHz.The intermodulation product raises the noise floor a significant amount, and would result in a reduction in the signal-to-noise ratio thus degrading the performance of the receiver.If the received signal is followed further through the front-end model of Figure 1, it can be seen that the two additional stages of filtering before the next gain stage will greatly suppress the lower-frequency interference signal preventing the generation of additional intermodulation products in the receiver LNA or mixer.The original intermodulation product, however, is now in the passband of the receiver and is not attenuated.
In laboratory testing of GNSS avionics in the presence of emulated interference from a proposed terrestrial broadband network, intermodulation products consistent with the theory described above were observed.Figure 11 shows the power spectrum of the emulated signals corresponding to the use of two Long Term Evolution (LTE) carriers in the 1525 to 1559 MHz band as shaped to emulate attenuation from filtering within an active antenna.Two originally proposed spectrum deployment phases were emulated.For phase 1 (blue), each carrier was modulated by a 5-MHz LTE signal.For phase 2, each carrier had a 10-MHz bandwidth.The details of the test setup are provided in Appendix A. Figure 12 shows the power spectrum of the LNA output voltage, and for comparison, the scaled power spectrum of the desired (GPS C/A-code) signal (with arbitrary power scaling to facilitate viewing).

| Reciprocal mixing
Traditional reciprocal mixing issues in communications receivers degrade receiver sensitivity in the presence of close-in interference signals. 10,12Phase noise and/or spurious signals (spurs) on the local oscillator (LO) signal of the receiver can translate the interference energy to the IF band of the receiver.While this is a particular concern for communications receivers where high level signals will occur in channels immediately adjacent to the operating channel, the same effect can mix interference signals at nearby frequencies into the IF band of a GNSS receiver.
Referring to the representative receiver block diagram of Figure 1, an ideal receiver will accept a signal at the receive frequency, f IN , mix it with an LO at f LO to generate an IF output signal at the difference between the two frequencies, f IF .If an interference signal that is just out-of-band appears at the receiver input and passes through the RF filters with little attenuation, it will also mix with the LO.Ideally, after being mixed to IF, this interference signal will fall outside the IF bandwidth and be rejected.This effect is illustrated in Figure 13.The mixer is a multiplier in the time domain.Thus, the mixer output spectrum may be computed as the convolution of the input RF and LO spectra in the frequency domain.If the frequency separation between the LO spur and the LO intended frequency is equal to (or close to) the frequency separation between the interference source and desired band, then after the mixer some of the undesired interference energy will combine with the desired signals in the IF band.
As a result of the convolution of signals within the mixer, the relative amplitude of the spur in the LO signal is maintained for the reciprocal mixed product.The resulting interference level at IF may thus be computed in a straightforward manner, as shown in the following example.Consider a high power interference source at f INT = 1555 MHz entering the antenna at −30 dBm along with an L1 GNSS signal at a power of −130 dBm.Assume the local oscillator is on the low side of the receive band and contains a spur at an offset of 20 MHz below the local oscillator frequency, at a relative level of −60 dBc.In this case, the spur will translate the interference source to the same IF that receives the GNSS signal, to generate an in-band IF interference signal.
Using the reference receiver diagram presented previously and assuming that the three pole filters reject the 1555 MHz signal by an additional 6 dB each, the signal and interference amplitudes at the mixer output can be calculated as shown in Table 1.
The mixer provides an effective +15 dB gain to the inband signal.However, since the LO spur responsible for the reciprocal mixed product is at −60 dBc, the interference signal experiences an effective gain of +15 dB to 60 dBc = −45 dB.The resulting I/S is around 16 dB.Even though the interference level used in the example is high, it will not saturate the front end but does cause a high degree of in-band interference through reciprocal mixing.
In reality, LO spurs are commonly generated as pairs of spectral lines in the LO generator and appear symmetrically about the LO signal.Even spurs that are not generated as part of the LO signal but are coupled into the LO signal path by some other means are commonly translated into symmetrical spurs when the LO signal is amplitude limited inside the mixer.So the effect of spurs at frequencies both above and below the LO signal need to be considered properly.Further, reciprocal mixing is not restricted to discrete spurious signals on the LO.Excessive phase noise on the LO will also mix with out of band spurious interference signals, potentially raising the noise floor at IF to degrade the signal-to-noise ratio.
The range of frequencies where reciprocal mixing is typically a concern occurs at offsets in the range of 10 to 50 MHz above and below the particular receive band.For L1, this is from 1525 to 1565 MHz and 1585 to 1625 MHz.For wide-bandwidth, precision receivers the susceptibility range is further increased by the wide IF bandwidth used within the receiver.

| Aliasing
Using results from other works, 13,14 the effects of out-ofband interference on receiver effective noise floor, (N 0 ) eff , including the effects of aliasing can be expressed as where f s is the receiver sample rate, S r f ð Þ is the power spectral density of the baseband filtered received signal, and S s (f ) is the power spectral density of the reference GNSS pseudorandom noise (PRN) signal.Thus, a GNSS receiver will have an increased sensitivity to interference concentrated on frequencies that will be folded, through the aliasing process, on top of the power spectrum of the desired signal.

| In-band emissions
Although the center frequency (or frequencies) of an outof-band interference source is, by definition, not in a GNSS frequency band, invariably, some of its transmitted energy will fall into the GNSS bands.Energy that is inband to GNSS can result from ( 1) the natural roll-off of transmissions that are at center frequencies immediately adjacent to GNSS bands, (2) harmonics at integer multiples of the interference source's center frequency due to nonlinearities in the transmitter, (3) intermodulation products at frequencies given by mf 1 ± nf 2 where f 1 and f 2 are two center frequencies of out-of-band sources, and (4) spurious emissions from the interference source due to various design imperfections.Harmonics and intermodulation products are often very difficult to control, since they may be generated externally to the transmitter. 15,16nterference from such mechanisms is best controlled through domestic and international spectrum management regulations.

| Additional filtering
The key to maximizing out-of-band interference performance is using sufficient filter rejection to minimize the level of interference sources so they cannot cause desensitization due to compression, IMD, or saturation issues in the active stages.A wide variety of filtering technologies are commonly used for GNSS receiver applications, which include the following:

| Dielectric resonators
Dielectric resonators are a very popular technology for GNSS RF and, occasionally, IF filters.These filters use small discs or cubes of low-loss high dielectric constant material as coupled microwave resonators to provide a low-cost, high-selectivity band-pass response. 5They are also often referred to as ceramic filters, since ceramic is a common dielectric material used in their fabrication.Ceramic filters tend to have a very well-behaved frequency response (eg, repeatable from unit to unit), very good bandwidth and rejection performance, and are preferred for lower volume high-performance receivers.

| Surface acoustic wave (SAW)
SAW filters are available for both RF and IF frequencies for GNSS receivers.SAW RF filters use resonators that operate by converting the input electrical signal into acoustic waves, using printed coupled resonant transducers, that propagate along the surface of a piezoelectric substrate. 17They are inexpensive and typically much smaller than dielectric resonators, which makes them an extremely popular choice for applications where size is of utmost importance, such as GPS receivers integrated into cellphones and other mobile devices.Unfortunately, there is little uniformity in the frequency response characteristics between different SAW filters which makes generalizing their performance difficult.SAW filters can also exhibit great variation in center frequency with temperature with a typical temperature coefficient of 30 ppm/ C. So, for instance, over the temperature range of −30 C to +85 C, the center frequency of an L1 GNSS SAW filter may vary by more than 5 MHz.Temperaturecompensated SAW filters have recently become commercially available with greatly reduced temperature coefficients but with increased cost due to the complexity of their fabrication.
At the lower IF frequencies, SAW filters use surface waves as time delay elements to achieve a finite element response (FIR) filter response similar to that resulting from digital filters.This is in contrast to implementing more traditional resonator style RF filters of higher frequency SAW devices.In FIR filters, amplitude and phase response are independent which allows simultaneous near-brick-wall frequency response and relatively flat group delay characteristic.In addition to the obvious benefits of passing the signal with little amplitude and phase distortion, the sharp transition region between the pass and reject bands suppresses interference signals which can be very close to the signal, which pass through the less selective RF circuits.Unfortunately, however, implementing the surface acoustic time delays necessary inside the device results in a large insertion loss as shown in the receiver model, as well as large absolute time delays between the terminals of the actual physical device.Further, small but finite reflections within the device will cause a small amount of amplitude and phase ripple to be measured on the device response.

| Bulk acoustic wave (BAW)
BAW filters 18 operate in a similar fashion to SAW filters in that they both operate through the use of resonators in which electrical signals are converted to acoustic waves.The difference between BAW and SAW filters is that, in BAW filters, the acoustic waves propagate through the substrate rather than along the surface before they are converted back into electrical signals.BAW filters have been gaining market share over SAW filters for mass-market RF applications because they can offer lower insertion losses and improved selectivity.BAW filter technologies include free-standing bulk acoustic resonators (FBAR) and solidly mounted resonators (SMR).BAW filters tend to exhibit less sensitivity to temperature (by about two-fold) than SAW filters, and temperature-compensated products are becoming available with near-zero temperature sensitivity.A principle BAW drawback with respect to SAW filters is that they are more difficult to manufacture and thus slightly more costly.

| Cavity filters
Cavity filters 5 offer low-insertion loss and high out-ofband attenuation, with their main drawback being that they are extremely large and heavy.They operate using similar principles as dielectric resonators, except that they utilize an air-filled cavity within a conductor rather than a dielectric block as the microwave resonator.

| Lumped component filters
Filters built using inductors, capacitors, and resistors are used at IF or baseband within many fielded GNSS receivers.Some lumped component filters that only utilize inductors and capacitors are referred to as LC filters, which follows from the common engineering symbols for inductance (L) and capacitance (C).][21] LC filtering suffers from filter response and group delay issues.It is difficult to obtain LC components with sufficient tolerance to ensure good response repeatability during manufacturing.They do have the advantage of better amplitude and phase ripple, and somewhat better insertion loss than SAW filters; however, they are not able to attenuate interference sources immediately adjacent to the pass-band as effectively as SAW IF filters.The amount of attenuation provided by such filtering depends on the design bandwidth of the LC filter and the order of the filter.As one example, Anon 20 describes a GPS L1 chipset that relies on a second-order, 15 MHz 3-dB bandwidth Butterworth LC filter centered at an IF frequency of 183 MHz.With this frequency plan, this filter provides ~10 dB of attenuation at 1550 MHz and ~40 dB at 1530 MHz.Active resistor-capacitor (RC) filters are also quite common in GNSS chipsets.These offer the benefit that they can be implemented internally to the chip, see, eg, Gramegna et al. 22

| Filter performance considerations
There is no single best filter technology for every GNSS application.Filters are generally selected based upon a careful consideration of numerous trade-offs between selectivity, insertion loss, differential group delay, size, weight, and cost.Definitions of these and other important filter characteristics are provided in Appendix B.
Figure 14 shows typical attenuation characteristics, and Table 2 presents a comparison of the typical characteristics of the various RF filter types identified above to illustrate some of these trade-offs.A more expansive treatment may be found in NPEF. 23As a caveat on interpreting typical performance characteristics, it is important to note that components can only be relied upon to meet specifications not "typical" values to allow for manufacturing and temperature variations.The relevance of specifications over typical performance is particularly true for filters with sharp cutoff transitions for which small variations can have very significant effects.The specifications allow for margin to account for such effects.

| Increasing linear region for active components
Interference-induced intermodulation distortion and compression effects are minimized by using amplifiers and mixers with sufficiently high compression points.Unfortunately, the highest compression points are accompanied by the highest operating current levels.Therefore, it is important to understand the interference environment as well as possible and model the front end on that basis.
It is possible through various design techniques to achieve greater linearity in a GNSS receiver's active components, see, eg, Morgan. 24Use of such techniques, however, comes at the expense of increased receiver noise figure and higher power consumption.In the end, careful system modeling and choice of appropriate amplifiers is the best strategy.

| Multi-frequency, multiconstellation considerations
Many GNSS receivers process desired signals from more than one GNSS constellation on two or more center frequencies.This functionality can be enabled by simply duplicating the full set of RF front-end components shown in Figure 1 for each center frequency, but this is rarely done.For instance, almost all multi-frequency GNSS receivers use just a single antenna (unless multiple antennas are needed for other reasons, eg, attitude determination).A broadband antenna or one that is resonant at multiple frequencies (eg, a stacked patch) may be used.Most multi-frequency receivers also share filters between desired passbands.A diplexer is a filter that passes two bands of frequencies, and a triplexer is used for three frequencies.LNAs may be shared between bands and aliasing may be intentionally exploited to share an A/D converter between two or more bands.Overall, for a multi-frequency GNSS receiver all of the same out-ofband interference coupling mechanisms must be considered as for a single-frequency receiver but mitigating performance degradations becomes more complicated.The designer must consider the impact of interference on the processing paths for each center frequency.

| Other design considerations
Good design practices help to optimize interference rejection.Power supply isolation is important to decouple the various RF stages from each other.While this is important for the stability of the active stages, it also prevents interference signals which appear at one amplifier from possibly bypassing the filters via the power supply lines to another amplifier farther down the line.The performance of filters which have 60 or 70 dB of ultimate rejection can easily by compromised by signals passing around them via the power supply lines.
There is a trend to open up the input bandwidth of A/D devices to very high frequencies.This makes the inputs susceptible to broadband noise and otherwise ignored interference sources.It is prudent to constrain the input bandwidth of the A/D to the chosen operating bandwidth.While a band-pass filter would work immediately prior to the A/D, it is often sufficient to simply cascade a two-or three-element low-pass filter with a similar order high pass filter to ensure that these unwanted signals are not aliased to frequencies which may degrade the input signal.

| Applicability to fielded equipment
It should be noted that there is extremely limited applicability of the mitigation approaches identified in this paper to fielded equipment.There may be some instances where additional filtering can be added to in-use systems, primarily for GNSS equipment that utilizes external active antennas.In such cases, it is possible that an existing antenna can be exchanged for a new one that employs additional filtering, or an in-line filter added between the external antenna and the receiver.Such modifications must be done only with extreme caution since they could detrimentally impact system performance in numerous ways.For a large quantity of fielded receivers, modification is not cost-effective.For instance, the receivers and antennas may be in an integrated package, in which case out-of-band interference issues can only realistically be resolved through replacement.

| SUMMARY
This paper has described the main mechanisms by which out-of-band interference can disrupt GNSS receiver functioning.These mechanisms include saturation and desensitization of front-end low noise amplifiers, mixers and other circuitry; reciprocal mixing effects that arise from the fact that receivers cannot generate a perfect tone to down-convert the desired signals; intermodulation products; aliasing of out-of-band emissions that remain after filtering into the receiver's passband; and the reception of in-band (to GNSS) emissions that are always present due to imperfections in the signal generation and filtering of the interfering system.
For most of these mechanisms, the best mitigation approach is through filtering properly distributed throughout the receiver front-end.Although there are a wide variety of filtering technologies available for use in GNSS receivers, not all are suitable for any given application.The trade-offs involved in selecting appropriate filters have been detailed.

APPENDIX A: | DESCRIPTION OF HARDWARE SETUP FOR INTERMODULATION EXAMPLE
Conducted emission tests of aviation receivers required the use of a filter/LNA that met RTCA DO-301 requirements.These tests introduce the combined GNSS signal and RFI source within the antenna assembly after the antenna element and just before the filter/LNA using either a power combiner or coupler.GNSS filter/LNAs built for aviation applications are normally integrated into an antenna assembly and not available as individual components.Furthermore, DO-301 requirements specify the overall frequency response, gain and P1dB for components in front of aviation receivers but do not dictate the exact implementation and therefore the distribution of filtering and gain is a choice that manufacturers make.A representative filter/LNA was constructed by using a combination of amplifiers and spectrally shaping the input interference signal that in effect achieved the requirements of DO-301 at the output of these components.Figure A1 displays the Mini-Circuits components used to achieve performance that met DO-301 requirements.Table A1 displays the pertinent characteristics of this combination of amplifiers, including the total gain, 1-dB compression point (P1 dB), third-order intercept point (IP3), and noise figure.Although the gain of this amplifier is much higher than required, the actual gain that was provided to individual aviation receivers was much lower due to cabling and a power splitter that provided test signals to multiple aviation receivers simultaneously.The net gain for any individual aviation receiver was +26 dB.
Interference signal generation was accomplished using an Arbitrary Waveform Generator (AWG) that had two independent channels and whose intended signals were created as complex representations of the desired signals centered on 0 Hertz.The two channels of the AWG were connected to a Vector Signal Generator (VSG) that translated these signals to RF.  A2 as they would appear at the antenna input.
The signals shown in Figure A2 would be filtered and amplified by components within an aviation antenna assembly and would therefore not be the same spectrally at the output of a DO-301 compliant filter/LNA.Figure A3 displays DO-301 filtering requirements and therefore the output from a DO-301 compliant filter/LNA must reduce spectral power to within these limits.
The desired frequency response at the output of the aviation LNA was achieved by spectrally shaping the broad band noise so that it included the filtering that DO-301 requires.Spectra of signals at the input of the representative aviation LNA are shown in Figure 11 and follow the frequency slope that DO-301 requires.

APPENDIX B: | GNSS RF FILTER PERFORMANCE CHARACTERISTIC DEFINITIONS
In this appendix, important performance characteristics for GNSS filters are presented.

FILTER TYPES
Band-pass and low-pass filters are most commonly used for GNSS applications.As illustrated in Figure B1, ideal band-pass and low-pass filters allow only a range of frequencies to pass through and completely suppress signals outside of this frequency range.For band-pass filters, the frequency range spans between two positive frequency values.Band-pass filters are used within GNSS receivers or active antennas to pass the desired RF GNSS signals before down-conversion (eg, at L1 ± 15 MHz for a receiver wishing to process all of the GPS L1 signals) and may also be used at IF after down-conversion.Low-pass filters are often found after down-conversion within receivers that utilize a low or no IF, or at times even with a high IF to suppress harmonics generated within the mixers used for down-conversion.

SELECTIVITY
Selectivity is the amount of attenuation that is provided by the filter towards undesired signals (ie, those that fall outside of the filter passband).Although as illustrated in Figure B1 ideal filters completely suppress signals outside of the passband, as will be seen in typical characteristics to be presented throughout this section, realizable filters can only provide a finite amount of attenuation.Achievable attenuation in realizable filters typically is generally small close to the passband and increases as frequency separation from the passband grows.
The passband is selected based upon the signals that are intended to be processed.For example, depicts the GPS L1 signals.Although 90% of the C/Acode power is contained within L1 ± 1 MHz, importantly, the ability of a GNSS receiver to precisely range upon the C/A-code or any other GNSS signal is enhanced tremendously in the presence of noise and multipath by additionally processing the sidelobes (see, eg, Van Dierendonck et al 25 ).Modern high-precision receivers generally utilize the full bandwidth of the signals transmitted by the GNSS satellites (40 MHz or more for some of the operational satellites) and often use a passband that is even broader than this for reasons related to group delay (see discussion below) and/or the desire to track signals that are or will be broadcast by multiple constellations (eg, GPS, GLONASS, Galileo, BeiDou, satellitebased augmentation systems [SBAS], QZSS).Many highprecision GNSS receivers utilize a common front-end to additionally process differential corrections that are transmitted by communication satellites in the 1525 to 1559 MHz band.
Selectivity is generally described using the magnitude of the transfer function,|H(f)|, of the filter, which is the ratio of the filter output spectrum to the spectrum of its input.Decibel (dB) units are typically used, with attenuation in dB equal to 20log 10 |H(f )|.For RF filters, vendors often provide scattering matrix parameters (S-parameters) with the S 21 parameter providing the transfer function.

INSERTION LOSS
Realizable analog filters will always provide some undesired attenuation of signals in the passband, which is referred to as insertion loss.Minimizing insertion loss is important, especially for any filter prior to the first significant gain stage within a GNSS receiver's front-end since this filter characteristic impacts the receiver's noise floor.In a benign environment, GNSS receivers see a noise floor that is due to 1. Undesired energy received from the antenna 2. Undesired energy from sources internal to the receiver/antenna, eg, due to thermal agitation of electrons within the antenna/receiver components.
Additional filtering will always increase the noise floor and the extent to which this occurs can be quantified using the expressions [(3)]: where N 0 is the noise density (in units of W/Hz) referenced to the output port of the passive antenna, k = 1.38E-23J/K is Boltzmann's constant, and T sys is the system temperature (in units of K).T s is the source or antenna temperature (75 to 100 K for a typical GNSS antenna that provides a broad gain pattern, ie, the upper hemisphere), T 0 is 290 K, and the entire second term in the right-hand-side of Equation (B2) is the receiver temperature.Of importance to the present discussion, the receiver temperature is influenced greatly by the first loss (with loss, in linear units, L 1 ) suffered between the output port of the passive antenna and the first gain stage (with gain, in linear units, G 1 ) in the receiver, as well as the noise figure, NF 1 , of the first amplifier.Losses and noise figures of components further downstream in the receiver front-end are of lesser importance if the gain, G 1 , is sufficiently high.Equation (B2) is recursive, in that its form repeats for additional losses, gains, and noise figures.
For the reasons described above, it is desirable for any filtering prior to the first low noise amplifier (LNA) within a GNSS active antenna or receiver front-end to have extremely low insertion loss.Insertion loss adds to the noise.Typical target design values can range from under 2.2 dB for aviation receivers with a clear view of the sky to less than 0.5 dB for some other applications.Filters with higher insertion losses can often be tolerated later within the RF/IF chain provided that the preceding net gain far outweighs the preceding net loss.So for instance, filters with insertion losses of up to 15 to 20 dB may be found at IF in numerous fielded receivers with little detrimental impact on receiver noise floor.

GROUP DELAY
The phase response of a filter is also of great importance for many applications.Any phase response within the passband that is not linear with frequency will distort the desired signals.The derivative of the phase response with respect to frequency is referred to as the group delay response because this function of frequency describes how much time delay is incurred upon each frequency component of the desired signal.
For navigation and positioning applications, the absolute value of the group delay is not consequential since it does not affect position accuracy.For such applications, the group delay differential, which describes how much the group delay varies over the passband, is the critical characteristic.It is desirable to keep the group delay differential as small as possible over the passband to enable better positioning performance.This design goal is especially important for receivers that make measurements from more than one satellite navigation system within a band, see, eg, Hegarty et al. 26 Receivers used for time transfer, ionospheric mapping, and other science applications also require the group and phase delays to be stable over variations in temperature.
Group delay differential generally grows with increasing filter selectivity, and the frequencies where maximum group delay differential is typically seen are in the transition region between the passband and stopband.For these reasons, many fielded high-precision receivers use passbands that extend beyond the 1559 to 1610 MHz radionavigation satellite services (RNSS) band.It is also possible for some filter technologies to use specialized designs to provide delay compensation to minimize group delay differential.The use of narrow filters with sharp cutoffs, as well as some implementations of delay compensation, increases the variation of group delay and phase versus temperature.
There exist many different ways to vary the gain of IF stages as necessary to maintain a suitable signal level atF I G U R E 2Typical local oscillator design the input of the A/D converter.Design choices for this function can have a profound effect on overall receiver susceptibility to interference.
receiver gain as a function of VGA voltage [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]

F I G U R E 5
Low noise amplifier system model F I G U R E 6 Input-output voltage characteristic for modeled airborne antenna LNA [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]F I G U R E 7 Input-output power characteristic for modeled airborne antenna LNA [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]

F I G U R E 8
Three-pole ceramic preselection filter attenuation (24 MHz 1-dB bandwidth, 33 MHz 3-dB bandwidth) [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]

F I G U R E 9
Input (top) and output (bottom) power spectrum of active antenna LNA; total interference power at antenna output port is −45 dBm, LNA input power is −57.5 dBm (LNA in linear region) [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]F I G U R E 1 0 Input (top) and output (bottom) power spectrum of modeled LNA; interference power at antenna output port is −25 dBm, LNA input power is −37.6 dBm (LNA in compression) [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org] However, if the LO contains spurious energy at some offset frequency Δf SPUR above or below the local oscillator frequency f LO , then out-of-band interference signals that appear at the input of the receiver meeting the criteria Δf SPUR =|f IN − f INT | will mix with this spur to generate interference signals which also fall inside the receiver IF bandwidth.

F I G U R E 1 1
Spectrally shaped broadband interference [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]F I G U R E 1 2 Intermodulation products seen in hardware testing [Color figure can be viewed ato wileyonlinelibrary.comand www.ion.org]

F
I G U R E 1 4 Representative attenuation characteristics of various GNSS RF filter types [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]

F
I G U R E A 2 Broadband interference signals [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]F I G U R E A 3 DO-301 antenna frequency response [Color figure can be viewed at wileyonlinelibrary.com and www.ion.org]

F I G U R E B 1
figure can be viewed at wileyonlinelibrary.com and www.ion.org] Two types of candidate interference signals were considered, each of which was below 1560 MHz.Both types consisted of broadband noise separated by a stop band.One interference signal, referred to as phase 1, consisted of two 5-MHz broad band noise signals occupying 1526.3 to 1531.3 MHz and 1550.2 to 1555.2 MHz.The other interference signal, referred to as phase 2, consisted of two 10-MHz broad band noise signals occupying 1526 to 1536 MHz and 1545.2 to 1555.2 MHz.Both of these intended interference signals are shown in Figure