GNSS Anti-CRPA System: Frequency Domain Filtering Principles That Actually Work in Jammed Environments

Spatial nulling gets all the attention in CRPA discussions. But frequency domain filtering is the workhorse that cleans up what spatial filtering leaves behind. A CRPA might suppress a jammer by 40 to 60 dB, but that residual energy still sits in the same frequency band as your satellite signal. You cannot beam it away. You have to filter it. Frequency domain filtering inside an anti-CRPA system is not a backup plan — it is a core layer of the suppression stack, and understanding its principles is what separates a receiver that survives jamming from one that just delays the inevitable.

Why Frequency Domain Filtering Exists Inside a CRPA Architecture

Spatial Nulling Is Not Enough

A CRPA array places nulls toward jammer directions. The null depth depends on array size, calibration quality, and algorithm convergence. Even a well-tuned 7-element CRPA running RLS achieves maybe 50 to 60 dB of null depth against a strong narrowband CW jammer. That sounds impressive until you realize the jammer might be 80 dB above the satellite signal. You still have 20 to 30 dB of jammer energy at the correlator input.

That residual energy is in-band. It overlaps the satellite signal in both frequency and code. A matched filter cannot separate them. The only way to reduce this energy further is to exploit any frequency-domain difference between the jammer and the signal. If the jammer occupies a narrower bandwidth than the signal, or if it sits at a slightly offset frequency, frequency domain filtering can carve it out without touching the satellite.

This is why every serious anti-CRPA system includes a frequency domain filtering stage after the spatial filter. The two layers attack the jammer from different angles — spatial filtering removes what it can by direction, and frequency filtering removes what it can by spectrum. Together, they push the residual jammer power below the tracking loop threshold.

Where Frequency Filtering Sits in the Signal Chain

The signal flows from the CRPA elements through the beamformer, then into the tracking loops. Between the beamformer output and the correlator, there is a frequency domain processing stage. This stage operates on the composite signal — satellite plus residual jammer — and applies spectral shaping before the code and carrier tracking loops see it.

The placement matters. If frequency filtering happens before the beamformer, it removes jammer energy that the adaptive algorithm could use to update its weights. The null would not converge properly because the algorithm never sees the full jammer power. If filtering happens after the tracking loops, the loops have already been corrupted by residual jammer energy and cycle slips have already occurred. The sweet spot is between beamforming and correlation — after spatial suppression, before tracking.

Core Frequency Domain Filtering Techniques

Notch Filtering Against Narrowband Jammers

Narrowband CW jammers are the easiest targets for frequency domain filtering. A CW jammer on GPS L1 occupies maybe 1 to 5 MHz of bandwidth. The GPS L1 C/A signal spans 2.046 MHz. If the jammer sits at 1575.42 MHz and the signal is centered at the same frequency, a simple notch filter cannot separate them. But most jammers are not perfectly centered. They drift by tens or hundreds of kilohertz due to oscillator instability.

A narrow notch filter — typically 2 to 5 MHz wide — centered on the jammer frequency removes most of the jammer energy while preserving the bulk of the signal spectrum. The notch depth of 30 to 40 dB is achievable with a simple IIR filter running at the sampling rate. The computational cost is negligible — a few hundred multiply-accumulate operations per millisecond.

The challenge is jammer frequency estimation. The filter must track the jammer frequency in real time. If the jammer hops or drifts, the notch must follow. A frequency-locked loop monitors the spectral peak and feeds the notch center frequency to the filter coefficients. This creates a tracking notch filter that follows the jammer across its frequency range.

Notch filtering is most effective against CW and slow-sweep jammers. Against fast-hopping jammers that dwell on each frequency for less than a millisecond, the notch cannot track fast enough and the filter becomes useless. For those threats, you need wider-band techniques.

Bandpass Filtering for Wideband Signal Protection

Wideband jammers sweep across multiple GNSS bands simultaneously. A notch filter cannot track a jammer that jumps from GPS L1 to Galileo E1 to BeiDou B1 every few milliseconds. The solution is bandpass filtering that preserves the signal band while attenuating everything outside it.

The GNSS signal occupies a known bandwidth — 2.046 MHz for GPS L1 C/A, 1.023 MHz for Galileo E1, and so on. A bandpass filter with a passband matched to the signal bandwidth and stopbands that attenuate out-of-band energy by 40 to 60 dB effectively removes any jammer energy that falls outside the signal band.

The problem arises when the jammer is wideband and covers the entire signal band. A swept jammer that dwells across the full 2 MHz of GPS L1 C/A cannot be removed by bandpass filtering because it is inside the passband. This is where sub-band processing becomes necessary — splitting the wideband signal into narrower chunks, filtering each chunk independently, then recombining.

Sub-band bandpass filtering is computationally heavier than a single wideband filter. A 7-element CRPA processing 8 sub-bands at 1 kHz update rate needs 8 parallel filter banks, each with its own coefficient set. But the suppression gain is worth it — each sub-band can have a notch or bandpass filter tuned to its specific jammer content, which is far more effective than a one-size-fits-all filter.

Adaptive Spectral Subtraction

Fixed filters work when the jammer spectrum is known and stable. In reality, jammers change. A pulsed jammer has a spectrum that varies with duty cycle. A swept jammer has a spectrum that moves across the band. A chirp jammer has a spectrum that changes shape continuously.

Adaptive spectral subtraction estimates the jammer spectrum in real time and subtracts it from the received signal. The estimation uses a reference channel that contains only jammer energy — typically a spatially nulled output from a beamformer element that points away from all satellites. This reference channel sees the jammer but not the satellite signal. Its spectrum is a clean estimate of the jammer's spectral shape.

The receiver subtracts a scaled version of this reference spectrum from the main channel. The scaling factor is updated continuously to minimize the residual power in the main channel. This is essentially a Wiener filter in the frequency domain, and it adapts to any jammer spectral shape — CW, pulsed, swept, chirped, it does not matter.

The limitation is reference channel quality. If the reference channel picks up any satellite signal — due to imperfect nulling or multipath — the subtraction removes part of the signal along with the jammer. This creates signal distortion that degrades tracking accuracy. The reference channel must be clean, which means the spatial nulling must be good enough that the reference sees negligible satellite energy. This ties adaptive spectral subtraction directly to CRPA performance.

Frequency Domain Filtering in Multi-Constellation Receivers

Handling Multiple Signal Bandwidths Simultaneously

A modern multi-constellation receiver tracks GPS L1 C/A, L5, Galileo E1, E5a, E5b, BeiDou B1, B2, and possibly GLONASS L1OF. Each signal has a different bandwidth, different center frequency, and different spectral shape. A single frequency filter cannot handle all of them.

The solution is per-signal filtering. Each signal gets its own frequency domain processing chain after the beamformer. The beamformer output is split into multiple down-converted channels, one per signal. Each channel applies filters tuned to that signal's bandwidth and center frequency. The filtered outputs feed independent tracking loops.

This per-signal approach lets you apply aggressive notch filtering to a jammed signal while leaving other signals untouched. If GPS L1 is being jammed but Galileo E1 is clean, the L1 channel gets a deep notch while the E1 channel runs unfiltered. The navigation solution fuses measurements from all constellations, so losing GPS L1 does not kill the position fix.

The computational cost scales with the number of constellations. A 6-constellation receiver with 3 signals per constellation needs 18 parallel filter chains. Modern DSPs handle this without breaking a sweat, but the power budget matters for embedded platforms. Filtering is often the second-largest power consumer after the beamformer itself.

Inter-Signal Filtering for Swept Jammers

Swept jammers that move across multiple constellations create a cross-signal interference problem. The jammer energy in one signal's band leaks into adjacent bands due to filter roll-off. A GPS L1 notch filter with 30 dB stopband attenuation still lets some energy through at the band edges. If the jammer sweeps close to the band edge, that leaked energy corrupts the adjacent signal.

Inter-signal filtering addresses this by monitoring the spectrum of each signal channel and applying cross-channel suppression when energy appears in adjacent bands. If the GPS L5 channel detects jammer energy near its upper edge, the filter tightens its roll-off and the adjacent Galileo E5a channel applies a preemptive notch at the overlapping frequency.

This coordinated filtering requires the channels to share spectral information in real time. The communication overhead is small — a few bytes per millisecond per channel — but the synchronization must be tight. If one channel updates its filter coefficients 10 milliseconds after the jammer arrives, the adjacent channel has already been corrupted.

Practical Constraints That Shape Filter Design

Group Delay and Its Impact on Code Tracking

Every digital filter introduces group delay — a frequency-dependent time shift in the signal. A sharp notch filter has high group delay near its edges. A steep bandpass filter has group delay across its passband. This delay distorts the code correlation peak.

The code tracking loop relies on precise timing. If the filter introduces 50 nanoseconds of group delay, the code phase measurement shifts by 50 nanoseconds, which is about 15 meters of range error. For precision applications, this is unacceptable.

Filter design must balance stopband attenuation against group delay. A filter with 40 dB stopband attenuation might have 20 nanoseconds of group delay. A filter with 20 dB attenuation might have only 5 nanoseconds. In many anti-CRPA systems, the filter is deliberately designed with moderate attenuation and low group delay, relying on the spatial nulling to provide the bulk of the jammer suppression. The frequency filter is a cleanup stage, not the primary defense.

Filter Coefficient Update Rate

The jammer spectrum changes over time. Filter coefficients computed 100 milliseconds ago are stale if the jammer has hopped or swept. The coefficient update rate must match the jammer dynamics.

For CW jammers, update rates of 10 to 50 Hz are sufficient. The jammer frequency drifts slowly, and the filter can track it with a simple PLL. For swept jammers, update rates of 500 Hz to 1 kHz are needed. The filter must recompute coefficients every millisecond or faster to keep up with the sweep.

For pulse jammers, the filter must react within microseconds of the pulse onset. This requires pre-computed coefficient sets for different jammer states, switched in real time based on a pulse detector. The detector triggers the coefficient swap, and the filter is already tuned to the new state before the pulse energy arrives.

This coefficient management system is one of the most underappreciated parts of frequency domain filtering. The filters themselves are simple. The logic that decides which coefficients to use and when to switch them is where the real engineering effort goes.

How Frequency Filtering and CRPA Nulling Interact

The Suppression Stacking Effect

Spatial nulling and frequency filtering do not just add their suppression gains — they multiply them in terms of effective jammer power reduction. A 50 dB spatial null followed by a 30 dB frequency notch gives you 80 dB of total suppression. That is not 50 plus 30. It is 50 plus 30 applied to the residual, which is a multiplication in linear power terms.

This stacking effect is why the two layers are both necessary. Spatial nulling alone cannot reach 80 dB. Frequency filtering alone cannot reach 80 dB. Together, they get there — and 80 dB is what you need when the jammer is 90 dB above the signal and you need 10 dB of carrier-to-noise ratio for the tracking loop to survive.

When One Layer Fails, the Other Must Compensate

In practice, the two layers do not always perform at their peak simultaneously. The CRPA null might degrade due to jammer movement or array calibration drift. The frequency filter might degrade due to coefficient staleness or group delay limitations. When one layer slips, the other must pick up the slack.

System designers account for this by setting minimum performance requirements for each layer independently. The CRPA must deliver at least 30 dB of null depth even under worst-case calibration. The frequency filter must deliver at least 20 dB of attenuation even with stale coefficients. If either layer falls below its minimum, the system triggers a degraded-mode alert and the navigation filter inflates its measurement noise accordingly.

This independent minimum requirement philosophy is what keeps the system honest. It prevents the common engineering trap of designing two layers that look great on paper but both depend on the same assumption — which means they fail at the same time.