Most SIGINT procurement failures are not hardware failures. They are specification failures. A vendor delivers exactly what the contract says, and what the contract says turns out to be insufficient to do the job. The system covers the right frequency range on paper but has a noise figure 8 dB worse than needed in practice. The direction-finding accuracy is "2 degrees" in the vendor's anechoic chamber but 9 degrees in the field conditions the system actually operates in. The software integration clause says "ATAK compatible" but the integration requires six months of custom engineering at additional cost.
Fixing these problems after contract award is expensive and slow. Fixing them before contract award requires writing specifications with enough technical precision that there is no room to misrepresent performance. This guide walks through every layer of a SIGINT system specification that matters — RF front-end parameters, geolocation accuracy, processing throughput, software integration, and acceptance test design — with the specific numbers and test conditions that make specifications enforceable.
Why SIGINT procurement fails
Procurement failures cluster around three causes. The first is vague performance specifications that allow vendors to select the most favorable test conditions. "Sensitivity of −110 dBm" means nothing without specifying the signal type, bandwidth, required probability of detection, and false alarm rate at that sensitivity. A vendor can meet that number at 433 MHz in a shielded lab while the system fails to detect the same signal at 900 MHz in a vehicle with a mediocre antenna.
The second cause is the absence of defined acceptance criteria. A specification that lists requirements but does not define the test procedure used to verify them cannot be enforced. Vendors know this. Without a contractually binding test plan, acceptance becomes a negotiation rather than a measurement.
The third cause is ignoring software integration costs in the total cost of ownership calculation. A SIGINT sensor that does not output data in formats your C2 system can ingest requires custom integration work. That work takes time, introduces failure points, and is rarely budgeted at the time of acquisition. Specifying data output formats and API requirements as mandatory — not optional — contract terms eliminates most of this risk at the specification stage.
RF front-end specifications
The RF front-end defines the fundamental observable space of the system. Parameters set here cannot be compensated for in software.
Frequency range
Specify the required coverage as a continuous range, not a list of channels or spot frequencies. A system that covers "VHF, UHF, and L-band" is ambiguous; a system that covers "30 MHz to 3000 MHz continuous" is not. If the threat set includes HF communications (3–30 MHz, used for long-range COMINT and skywave propagation scenarios), microwave relay links (6–11 GHz), or UAV data links (5.8 GHz, Ku-band), extend the range accordingly and verify that sensitivity specifications apply across the full range, not only at mid-band.
Require the vendor to provide a measured sensitivity figure — not a nominal figure — at a minimum of five frequency points distributed across the full coverage range, with the measurement methodology specified (signal type, bandwidth, detection threshold, false alarm rate). Sensitivity that degrades by more than 6 dB from the stated figure at any point in the operational band should be a contract non-conformance.
Noise figure and sensitivity
Noise figure (NF) is the fundamental measure of receiver front-end quality. A receiver with NF of 5 dB is 5 dB more sensitive than the theoretical minimum; a receiver with NF of 15 dB has 10 dB less sensitivity — which is the difference between detecting a signal at 10 km and detecting it at 3 km. For tactical SIGINT applications, specify NF of 8 dB or better across the primary coverage band. Pre-amplified antenna inputs can reduce the effective system NF to 3–5 dB for applications requiring maximum sensitivity on weak signals.
Minimum detectable signal (MDS) translates NF into a practical sensitivity floor: MDS = −174 dBm + NF + 10·log10(bandwidth). At 25 kHz receiver bandwidth and NF of 8 dB, MDS is approximately −122 dBm. Specify MDS explicitly rather than relying on NF alone, because MDS is directly testable with a calibrated signal generator at a defined SNR threshold and probability of detection.
Instantaneous bandwidth, dynamic range, and ADC resolution
Instantaneous bandwidth (IBW) determines how much spectrum is captured simultaneously. A 25 MHz IBW receiver monitors a 25 MHz slice; a 100 MHz IBW receiver monitors four times as much simultaneously without frequency hopping. For applications requiring monitoring of the full tactical communications band without gaps, specify IBW of at least 40 MHz for VHF/UHF coverage. Higher IBW increases processing load proportionally — ensure the processing specifications account for the aggregate sample rate the chosen IBW implies.
Dynamic range, expressed as spurious-free dynamic range (SFDR) and third-order intercept point (IP3), determines whether the receiver can handle strong nearby signals without generating intermodulation products that mask weak signals of interest. Specify SFDR of at least 80 dB and IP3 of at least +10 dBm for tactical environments where co-site strong transmitters are common. Receivers with inadequate dynamic range produce phantom signals — detected signals that are actually intermodulation products of real signals — which are extremely difficult to identify and filter during operations.
ADC bit depth sets the digitization precision. 14-bit ADCs are the practical minimum for tactical SIGINT; 16-bit ADCs provide the dynamic range needed for challenging co-site environments. Vendors sometimes advertise high-bit-depth ADCs but achieve effective number of bits (ENOB) several bits lower due to clock jitter and thermal noise. Require the vendor to state ENOB as well as nominal bit depth, and specify ENOB of at least 12 bits across the primary coverage band.
Geolocation specifications
Emitter location is one of the highest-value SIGINT products. Specifying geolocation performance requires separating single-site and multi-site capabilities, because they have fundamentally different accuracy limits and error sources.
Single-site direction finding
Single-site DF produces a bearing — an azimuth from the sensor to the emitter — not a position fix. Accuracy is expressed as bearing RMS error in degrees. Specify bearing RMS error as a function of SNR and terrain condition. A reasonable requirement for a competent 8-element circular array is 2 degrees RMS at SNR above 20 dB in open terrain, degrading to no worse than 6 degrees at SNR of 10 dB. Require measurements at a minimum of 36 test azimuths (every 10 degrees) to expose array calibration errors and pattern asymmetries that are invisible in cherry-picked test scenarios.
The DF algorithm also matters. Correlative interferometry and MUSIC-based superresolution algorithms outperform simple phase comparison at low SNR. Watson-Watt algorithms are fast but less accurate in multipath. Specify the required algorithm class if the SNR environment is known, or require the vendor to demonstrate performance across multiple algorithm options.
Multi-site TDOA and FDOA geolocation
Time-difference of arrival (TDOA) geolocation combines bearing or time-difference measurements from two or more geographically separated sensors to compute a position fix. Accuracy is expressed as circular error probable (CEP) — CEP50 means 50% of fixes fall within that radius of truth, CEP90 covers 90%. Specify both CEP50 and CEP90 to characterize the tail of the error distribution, which matters for operational planning. A system with good CEP50 but poor CEP90 has occasional large outlier errors that can send forces to the wrong location.
TDOA accuracy depends on timing synchronization precision across sites. GPS-disciplined oscillators achieving 100 ns timing accuracy are the practical standard; specify the required timing synchronization accuracy in the procurement document and require the vendor to show how it is achieved and verified. Cross-correlation peak width is a function of signal bandwidth — wideband signals yield sharper TDOA estimates — so specify minimum signal bandwidth requirements for geolocation activation.
Frequency difference of arrival (FDOA), also called differential Doppler, is useful for mobile emitters when TDOA alone is ambiguous. Require FDOA capability if the operational scenario includes significant emitter or platform motion. Specify minimum relative velocity sensitivity for FDOA activation.
Processing throughput specifications
Processing specifications are where the most misleading vendor claims appear. Raw numbers — "classifies 500 signals per second" — are meaningless without the context of what fraction of the spectrum is covered, what the dwell time per signal is, and what the end-to-end latency looks like.
Collection duty cycle
Collection duty cycle is the fraction of time the system is actually sampling and processing the required frequency band. A system with 50% duty cycle on the primary band misses half of all signals, including those that transmit in short bursts. Specify minimum duty cycle of 95% or better on the primary coverage band for continuous-monitoring applications. For frequency-agile scanning applications, specify the maximum scan cycle time and the dwell time per channel, and verify these numbers with a frequency counter or spectrum analyzer during acceptance testing.
Latency from collection to analyst
The time from signal onset to analyst notification determines whether the intelligence is actionable. For time-sensitive targeting, specify end-to-end latency of under 5 seconds from first sample to alert delivery. This budget must cover signal detection, classification, geolocation computation, database insertion, and watchlist matching. Vendors who decompose the system into a chain of components may meet each component latency individually while missing the end-to-end requirement. Test latency end-to-end with a stopwatch and a known test signal — not by summing vendor-provided component estimates.
Emitter dwell time and classification confidence
Specify the minimum dwell time required for a reliable classification output. A system that requires 500 ms of dwell time to classify a signal will miss frequency-agile emitters that transmit in 20 ms bursts. Short-dwell classification requires either a very short observation window architecture or a separate burst-detection mode. Specify the minimum classifiable dwell time and the required minimum classification confidence score at that dwell time.
Software integration requirements
A SIGINT sensor that cannot share data with existing systems in the operational architecture has limited operational value, regardless of its RF performance. Data integration requirements must be specified as mandatory features, not as optional extras subject to separate negotiation.
Cursor-on-Target and ATAK output
Cursor-on-Target (CoT) is the XML schema used by ATAK and most Western tactical C2 systems to share position and track data. Require the system to output emitter position fixes and tracks as CoT events over UDP multicast at a configurable broadcast interval. Specify the required CoT event schema version and the mandatory fields (uid, type, time, stale, how, lat, lon, ce, le, hae). A system that outputs "CoT-compatible" data but omits the confidence fields (ce, le) or uses non-standard type codes will not display correctly in ATAK without customization.
MISP IOC export
For integration with threat intelligence workflows, require structured export of detected emitter parameters as MISP attributes. RF observables — frequency, modulation type, emitter fingerprint — are increasingly represented in MISP as custom object templates. Specifying MISP export enables detected emitters to be correlated against shared threat intelligence databases and feeds detection data into broader intelligence fusion workflows without manual data entry.
STANAG data formats and API
For programs operating within alliance structures, specify relevant STANAG format compliance. STANAG 4559 covers ISR tasking and collection management; compliance enables machine-to-machine tasking from a collection management system without operator intervention. STANAG 4609 covers motion imagery metadata that includes geolocation fields. Beyond standardized formats, require a documented REST API with authentication, versioned endpoints, and a written interface control document (ICD). The ICD should be a contract deliverable, not a promise to provide documentation after integration.
Test and evaluation design
The specification is only as good as the acceptance test that verifies it. A test plan designed by the vendor, conducted by the vendor, and reported by the vendor does not provide independent verification. Structure acceptance testing so that the procuring organization controls the ground truth and independently measures the results.
Signal generator test setup
For sensitivity and classification testing, use a calibrated signal generator (Rohde & Schwarz SMBV100B or equivalent) connected to the system antenna input through a calibrated attenuator. This provides repeatable, known signal levels across the full frequency range. Test sensitivity at each of the specified band points with the signal at MDS, 10 dB above MDS, and 20 dB above MDS. Record the probability of detection and false alarm rate at each level. Do not accept sensitivity data from vendor-run tests on signals of unknown level injected from an external antenna — these cannot be independently verified.
Ground-truth DF and geolocation scenarios
For DF accuracy testing, place a signal source at precisely surveyed azimuths from the antenna array. Use a total station or differential GPS to determine the true azimuth to the source to 0.1-degree accuracy. Transmit a known waveform at a defined power level and record the system's bearing output for at least 100 independent bearing estimates at each test azimuth. Compute the RMS error from truth. Test at a minimum of 8 azimuths distributed evenly around 360 degrees and at three range distances to verify that accuracy degrades gracefully with distance.
For TDOA geolocation testing, place the transmitter at a precisely surveyed location and compare the system's computed position fix against truth. Run at least 50 independent position fixes and compute CEP50 and CEP90 from the resulting position scatter. Test at multiple transmitter locations within the operational coverage area — accuracy varies with geometry (GDOP), and a single test location can hide poor performance at unfavorable geometries.
Integration and latency tests
End-to-end latency testing requires an independent timing measurement. Trigger a test signal at a known time using the signal generator's trigger output connected to a timestamp logger, and record the time at which the alert appears in the operator interface or is delivered over the data API. The difference is the true end-to-end latency. Run this test 50 times and report the mean, 90th percentile, and maximum latency — the 90th percentile and maximum are more operationally relevant than the mean.
For integration testing, connect the system to a representative instance of the target C2 environment and verify that CoT events appear correctly on the ATAK map, that API queries return correctly formatted responses, and that MISP exports parse without errors into a test MISP instance. Document all integration tests in the acceptance test report with pass/fail criteria defined before testing begins.
Corvus Intelligence supports defense and government procurement teams with independent technical specification review, vendor evaluation, and acceptance test design for SIGINT and RF intelligence systems — so your contract terms match operational requirements before contract award.
Explore Corvus Intelligence →