1The station as a chain, and where each fault lives

The transmit path runs from the modem, along the inter-facility link, into the block upconverter whose local oscillator is locked to the station's 10 MHz reference, through waveguide to the orthomode transducer, the feed and the reflector. The receive path mirrors it through the low-noise block. Each fault class has a location in the chain and a characteristic signature, and Figure 1 draws both chains with the points where a field analyzer connects.

Modem statistics (C/N, error counts, MODCOD changes) are the first indication of a fault and should be trended. They do not, on their own, locate it: a wet connector, a drifting reference, feed misalignment and adjacent-satellite interference all present as reduced C/N. The measurements that follow localize the fault to an element of the chain, and Table 1 maps each element of Figure 1 to the signature it produces when it fails.

Interference follows allocation, so Table 1 reads differently in each band. C band pairs a 5.925 to 6.425 GHz uplink with a 3.7 to 4.2 GHz downlink, X pairs 7.9 to 8.4 with 7.25 to 7.75, Ku pairs 14.0 to 14.5 with 10.95 to 12.75, and Ka pairs 27.5 to 31.0 with 17.7 to 21.2, the operator's band plan governing at the edges. Two of those allocations are crowded: the cleared 3.7 to 3.98 GHz segment in the United States, and the 13.75 to 14.0 GHz uplink segment shared with radiolocation. Both return in section 9.

2Commissioning begins with the site survey

A survey made before the reflector arrives is the least expensive measurement in this paper. It establishes what the site receives before the station has transmitted, and it is the reference that later shows whether an emitter predates the station. It is reported in field strength, because a coordination limit is written in field strength and a reading in dBm depends on which antenna and cable were used:

So a −60 dBm reading through that horn and cable is 47 dBµV at the connector and 79 dBµV/m in the air. The instrument applies the same correction point by point from an antenna factor table and a cable loss table, as Figure 2 shows.

Figure 2 is one sweep at one azimuth, and a survey is many of them. A sweep shows what is present while it runs; an intermittent emitter needs an observation over time. Figure 3 shows the form a survey record takes: max hold shows that something occurred, the spectrogram shows when, and the single sweep shows only what was present for the milliseconds it took.

1. Fix the geometry first: planned azimuth and elevation, the horizon profile, and every known terrestrial transmitter within line of sight, in the file before any RF is measured.
2. Sweep the receive band and the segments beside it at the look angle, both polarizations, max hold for at least ten minutes each, with azimuth, elevation, antenna, factor table, cable loss, resolution bandwidth, detector and time recorded.
3. Sweep the horizon in azimuth steps of half the survey antenna's beamwidth, with max hold long enough to catch a rotating radar at least twice; section 9 derives how long.
4. Watch the band for an hour or more with a gap-free capture, so anything intermittent appears with its period and duty cycle rather than as a single spike.
5. Archive everything with the instrument's model, serial, calibration date and the weather. A survey that cannot be reproduced cannot be compared with a later one.

3The link budget, and which terms can be measured on site

The satellite link equation is short, and few of its terms are under the station's control:

At a representative geostationary slant range of 38,500 kilometers, free-space loss is 196.2 dB at 4 GHz, 205.7 dB at 12 GHz and 210.2 dB at 20 GHz. Free-space loss at Ka exceeds C band by 14 dB over identical geometry, and for a fixed reflector the antenna gain at each end rises by the same 14 dB, because gain rises as the square of frequency. What Ka cannot recover is rain, which moves by tens of decibels in minutes, so Ka systems run adaptive coding. Table 2 works a downlink budget through line by line and marks the lines the station can measure.

One noise-floor check decides whether the analyzer is ever the limit on Table 2. A 150 K system is −174 + 10 log(150/290) = −176.9 dBm/Hz at the LNB input and, with 60 dB of LNB gain, −116.9 dBm/Hz at the L-band output. An analyzer with a published displayed noise level of −159.9 dBm/Hz at 1 GHz, in 1 kHz resolution bandwidth at a −50 dBm reference level, is 43 dB below that, so at the L-band port the instrument does not limit the reading. The one setting that can spoil a reading there is the reference level, and section 12 returns to it.

The uplink chain, and the reading at the coupler

Radiated power is not measured directly; it is assembled. A 4 W block upconverter is +36.0 dBm, or +6.0 dBW. The waveguide and feed after it lose 0.3 dB, by the manufacturer's figure and the sweep of section 8. A 1.8 m reflector at 60 percent efficiency has 46.4 dBi of gain at 14.25 GHz. The EIRP is therefore +52.1 dBW. The one number a field analyzer measures directly is the upconverter output at a −30 dB directional coupler, and it has two traps. Arithmetic: the coupler and 1.5 dB of test cable put the display 31.5 dB below the BUC output. Safety: a 400 W amplifier at +56 dBm puts +26 dBm on the same coupler port, above a +23 dBm maximum input, so a high-power chain needs a pad and the pad goes into the correction too. An amplitude offset entered in the instrument folds both into every point of the trace, noise floor included, so the screen reads power at the flange and not at the connector. The offset value is part of the record: two EIRP figures are comparable only when both name the reference plane they were taken at.

4Figure of merit, measured with the sun

Figure of merit is the receive side: antenna gain divided by system noise temperature, measured in the field by pointing first at cold sky and then at the sun, at the same elevation, and taking the ratio of the two noise powers. That ratio, conventionally called Y, gives:

The eight is 4π from the aperture Gλ²/4π and a two because a single-polarization feed collects half of an unpolarized sun. The correction L dominates the result on a large aperture, so it is derived here. For a Gaussian main beam of half-power beamwidth θ_hp and a uniform solar disc of diameter θ_s, the reciprocal of the fraction of the disc's power the beam collects is:

For a sun well inside the beam the expression reduces to 1 + 0.35x, the linear correction most handbooks print with a coefficient between 0.35 and 0.38, and Figure 4 shows where the two forms part company. Two idealizations are granted here. The sun at centimeter wavelengths is brighter at its limb than at its center, which puts more of its power where the beam is weaker and pushes the true correction up. A real main lobe is not Gaussian at its base and carries near sidelobes, which collect part of an overfilling disc the Gaussian ignores and pull it down. Both are small while the sun is inside the beam, and both are why handbook coefficients differ in the second decimal.

A worked case. A 1.8 m Ku station measured at 12 GHz has a half-power beamwidth near 70λ/D = 0.97 degrees, so the sun underfills it: x = (0.53/0.97)² = 0.30 and L = 1.11. With a measured rise of 11.5 dB between cold sky and the sun, Y = 14.1, and with 400 solar flux units at 12 GHz, equation (3) gives 3.470 × 10⁻²² × 13.1 × 1.11 / (6.241 × 10⁻⁴ × 4 × 10⁻²⁰) = 202, or 23.0 dB/K, the figure Table 2 assumed. Check it forward from the physics, not backward from the answer: a 1.8 m reflector at 60 percent efficiency has 44.9 dBi of gain at 12 GHz, and 44.9 dBi over 150 K is 23.1 dB/K. The two agree, and the sun raised the antenna temperature by about 2,000 K.

On a 9 m station the correction takes over, as Figure 4 shows. Its beam is 0.19 degrees, so the sun overfills it nearly three times over: x = 7.4 and L = 5.19. A measured rise of 18.6 dB gives Y = 72.4, and 3.470 × 10⁻²² × 71.4 × 5.19 / (2.496 × 10⁻²³) = 5,150, or 37.1 dB/K. Forward: 58.9 dBi over 150 K is 37.1 dB/K, and the antenna temperature on the sun was about 10,700 K, the order of the sun's brightness temperature at this wavelength.

Three practical notes. The flux must be at the measurement frequency: observatories report at a few fixed frequencies, 8.8 and 15.4 GHz being the nearest to Ku, and the 12 GHz value is interpolated on a log-log line; that is how 400 units was reached from 270 and 560. The measurement is made above about twenty degrees of elevation, so the ground stays out of the sidelobes, and repeated per polarization with the spread reported. And the Y-factor is written down with its date, because the rise measured on purpose is the depth of the sun outage of section 8.

5Line-up and carrier access, and what goes in the file

Commissioning is a sequence, and each step produces a dated record that later measurements are compared against. The sequence runs: point and peak, confirm the satellite, null the cross-polarization under the operator's direction, step the power up, bring modulation up and measure it against the mask, then archive.

Pointing and peaking

Peaking is done on the satellite's beacon with the analyzer on the L-band monitor port, in a narrow span, by moving the antenna through a small angle in each axis. A max-hold trace held against the live trace makes it repeatable: the antenna is on axis when the live peak meets the max-hold peak and no further movement raises it. Zero span with the sweep time set to the beam crossing draws the peaking curve directly, in the form to file; a flat-topped or asymmetric curve indicates feed defocus or reflector distortion. Confirm the satellite by matching the observed carrier plan against the operator's, not by signal strength, because a strong beacon on the wrong satellite peaks in the same way. Record the beacon level with the date and the weather as the pointing baseline.

Cross-polarization, power and modulation

Under the operator's direction, transmit a clean carrier at low power while they measure the co-polarized and cross-polarized levels, and rotate the feed to the null. Operator access standards state a minimum discrimination, commonly about 30 dB, and the figure in the file is the one the operator measured. Re-check after final torque-down, because the null moves when the feed is clamped. Then step to the assigned radiated power under supervision, watching the shoulders at the coupler at each step, bring modulation up, and measure the occupied bandwidth and the spectral shape against the operator's mask. Occupied bandwidth is part of the access record, and so is its method: a −3 dB width and a 99 percent power width of the same carrier are different numbers, and the operator's procedure says which it wants.

The mask record is the transmit spectrum against the operator's mask, with the resolution bandwidth, video bandwidth, detector, reference level and offset recorded alongside it: a wider resolution bandwidth fills the shoulders in and a peak detector raises them. The third record is channel power at the coupler, integrated over the declared bandwidth with an RMS detector at the authorized drive, which is the station's EIRP record; a peak-detected maximum overstates a modulated carrier by several decibels.

The archive, and what it is for

Figure 5 lists the file row by row, with the later comparison each row supports.

Figure 5 summarizes the paper's argument, and it sets two conditions on the instrument on site: its readings carry a published accuracy with a stated condition, and its records carry their settings. A field analyzer that exports every record with its settings attached meets the second; whether its published figures meet the first is the question of sections 9 and 10.

6Transmit chain health: back-off and intermodulation

Back-off is the least expensive fix in the chain, and it has a cost. Every decibel of output back-off is a decibel off the link's carrier-to-noise, and Table 2 shows a 19 dB clear-sky margin that rain will take most of. Third-order products fall three decibels for every decibel of output back-off while the carrier falls one, so one decibel of back-off improves the shoulder-to-carrier ratio by about two decibels at the cost of one decibel of link margin. The two-tone test prices it: two equal tones produce third-order products at twice one frequency minus the other, and their level below the tones gives the intercept point:

One further check belongs with the arithmetic of Figure 6, because the analyzer is itself nonlinear. The ICX-400R publishes an input intercept of +27.4 dBm at 1 GHz with the reference level at 0 dBm, the setting of this capture, and +40.3 dBm with it at +20 dBm. At 0 dBm the analyzer's own third-order products from −6.0 dBm tones would lie 2 × (27.4 + 6.0) = 66.8 dB below the tones, 5.8 dB above the −72.55 dBc measured. The published figure therefore cannot separate the analyzer's products from the device's in this capture; the reading shows only that this unit did better than its published bound. Two steps settle it. Measure with the reference level at +20 dBm, where the published intercept puts the analyzer's products 2 × (40.3 + 6.0) = 92.6 dB below the tones, 20 dB under this reading. And add 10 dB of input attenuation: products that fall by 20 dB relative to the tones were generated in the analyzer, and products that do not move belong to the device.

Harmonics matter for a reason the mask does not cover: the second harmonic of a C-band uplink near 6 GHz lands near 12 GHz, inside the Ku downlink band, and C-band waveguide passes it unless the chain carries a harmonic filter. In a harmonics table, a row whose frequency is not an integer multiple of the fundamental is a peak search returning the noise floor, recorded as a bound, not a value. A harmonic that was 60 dB down at commissioning and is 40 dB down a few years on says the amplifier has moved toward saturation, which the carrier power alone cannot show.

7Frequency references and phase noise

A Ku block upconverter locked to the station's 10 MHz reference multiplies it by 1,305 to reach a 13.05 GHz local oscillator, and phase noise rises with the square of the multiplication:

A reference whose phase noise is adequate at 10 MHz can limit modulation order at 14 GHz, because the multiplication adds 62.3 dB. The symptom is a modem that locks at low-order modulation and fails at high order while every level on the analyzer reads correctly, because phase noise moves no power. The measurement to take before the block upconverter is examined is the reference's own phase noise; Figure 7 shows a phase-noise measurement on a 1 GHz source.

Figure 7 is read against the instrument's own published figures. The trace runs within 1.6 dB of the published typical phase-noise measurement floor of the 9 GHz handheld models at every offset from 1 kHz to 1 MHz (−117, −124, −129 and −139 dBc/Hz), so it is consistent with a reading at or near an instrument floor. It is used here for its shape and as an upper bound on the source, not as the source's level. For the instrument recommended in section 10, the published phase noise at 1 GHz and 10 kHz offset is −107.5 dBc/Hz, at 25 °C after a 10 minute warm-up, and that is the figure to plan against. The comparison illustrates a rule for any analyzer: a reading within about 10 dB of the instrument's own figure measures the instrument as much as the source, and such a reading wants a second method.

The figure also shows how the penalty scales with where the chain starts. Carried from 1 GHz to 13.05 GHz the multiplication is 13.05 times, or 22.3 dB, so a source at this trace arrives at no worse than −102.6 dBc/Hz at 10 kHz offset. From a 10 MHz reference the multiplication is 1,305 times, or 62.3 dB, so a reference at −120 dBc/Hz at 10 kHz arrives at −57.7 dBc/Hz, and the integration below shows what that costs a high-order constellation.

The law has a boundary: a converter multiplies its reference inside a phase-locked loop, so the law holds inside the loop bandwidth, typically hundreds of hertz to a few kilohertz, and outside it the converter's own oscillator sets the floor. The reference governs the close-in noise; the converter's datasheet governs the rest. What limits a modem is the integral of the curve rather than the curve itself: a carrier-recovery loop tracks phase noise inside its bandwidth, and the noise outside it becomes a random rotation of every symbol:

Take the trace of Figure 7 as the source's phase noise, which overstates it if the trace is an instrument floor, and integrate it from 10 kHz to 1 MHz on log-linear segments between the tabled points: 4.9 × 10⁻⁸ rad², so φ_rms is 0.31 milliradians, or 0.018 degrees. Multiplied by 13.05 to the Ku local oscillator that is 0.23 degrees rms, about 0.4 percent of error vector magnitude. Multiplied by 1,305 from a reference with the same curve, through a loop wide enough to pass it, it would be 23 degrees rms. A rotation of 7.7 degrees carries the outer corner of a 64-QAM constellation, at (7, 7) on its unit grid, across the decision boundary at 6, so the first case leaves more than thirty times its rms error in margin and the second exceeds the margin threefold. Figure 8 is what the first case looks like on a demodulator.

Figure 8 is a bench carrier; at an earth station the same measurement is made at the L-band monitor port on the station's own carrier, where the phase error column reads the whole chain's reference and converter contribution at once. The 710 parts per billion offset in the phase-noise capture is the other thing to read: at 13.05 GHz it is 9.3 kHz, and a receiver's acquisition range is a specification with a number in it. Judging it needs a timebase better than the thing judged: 1 ppm at 14.25 GHz is 14 kHz of uncertainty, which cannot resolve a 10 kHz offset, while 0.15 ppm is 2.1 kHz, which can. The ICX-FieldHawk-R publishes a frequency accuracy under 1 ppm with its standard TCXO and with the option 01 OCXO alike; the OCXO's 0.15 ppm is its temperature stability, with aging under 1 ppm a year, both at 25 °C after a 10 minute warm-up. The 0.15 ppm is therefore reached by locking the analyzer to an independent 10 MHz reference better than the station's own, through the external reference input or the option 06 GNSS-disciplined reference clock, and option 01 holds that setting over temperature. That is a decision made on the order form, not on the day.

8Preventive maintenance: the calendar and the ladder

Maintenance is the commissioning file, repeated on a schedule, with each new record compared against the one before, and much of it is spent at the joints of the RF chain, where water reaches a transmission line first. The cadence is set by how fast each part degrades and how much station time the measurement costs: a loop test takes a minute and runs before every pass, a figure of merit takes two hours and runs each quarter. Table 3 is the calendar, and the record column is the point of it.

The receive-side rows of Table 3 use four tests, listed in Table 4 in order of the station time they cost, each with the span of the chain it proves and the part it cannot. A failed rung therefore names the elements still in question before anyone climbs the structure.

Cable and waveguide, by scalar sweep

When the RF loop test of Table 4 fails, the inter-facility cable is the element it covers that is easiest to test, and the annual sweep is how a cable is checked before it fails. Insertion loss should rise smoothly with the square root of frequency; a wet connector adds disproportionate loss at the high end and also puts periodic ripple on the trace. That ripple gives the distance to the fault:

The stimulus is a Signal Generator swept across the band into one end of the run, with the analyzer at the other end in max hold. A reference sweep through the test cables alone is subtracted. Return loss needs a directional bridge or coupler at the source end, and in dispersive waveguide the distance uses the group velocity at the measurement frequency. The numbers to carry: a return loss of 20 dB is a reflection coefficient of 0.10, a VSWR of 1.22 and a mismatch loss of 0.04 dB, a healthy field joint; 14 dB is 0.20, 1.50 and 0.18 dB, marginal; and below 9.5 dB, where the VSWR passes 2:1, something is mechanically wrong.

The annual visit carries three checks the sweep does not. The DC check: the inter-facility cable usually carries the converters' supply, corrosion raises its loop resistance over years, and a converter that resets under load is a cable fault no sweep shows, so measure the voltage at the far end under load. The flange inspection: every bolted joint, gasket, pressure window and drain, because a sweep does not show a corrosion track. And the installation discipline: cables labeled at both ends, signal separated from power, cables on supports so water runs beneath them, and one grounding grid that everything reaches.

Rain fade and sun outage

Two losses are scheduled by physics and not by the calendar, and both belong in the plan so they are not reported as faults. Rain attenuation at Ku and Ka is predicted from the site's rain rate and answered by adaptive coding and modulation; the record is the fallback log read against the rain gauge, and a fade that does not match the weather is a fault.

The sun outage can be predicted to the minute. Twice a year, near the equinoxes, the sun passes behind the satellite, and for a few minutes on each of a few consecutive days its noise fills the beam. The depth of the outage is the Y-factor of section 4, the same geometry seen from the other side. The 1.8 m station that measured an 11.5 dB rise on the sun loses 11.5 dB of carrier-to-noise at the center of its outage, which takes the 19.0 dB of Table 2 to 7.5 and drops every high-order carrier. The sun crosses the sky at a quarter of a degree a minute, so the transit through a 0.97 degree beam plus the sun's 0.53 lasts about six minutes, and the 9 m station's 0.19 degree beam is crossed in under three. The operator publishes the dates; the station's task is to have them in the log before the alarm.

9Interference, and matching observation time to the interferer

Continuous interference is found by a sweep: adjacent-satellite carriers, cross-polarization, a terrestrial link in the band, the station's own intermodulation products. Point an analyzer at it and it is there. Intermittent interference takes longer to resolve, and against it a short swept survey is unreliable unless the observation time is matched to the interferer's repetition period. Table 5 sorts the candidates by mechanism, and the arithmetic that follows gives the observation time the last two rows require.

The fourth row of Table 5 is worked here because its mechanism leaves nothing visible in band. A full Ku transponder from a satellite at +48 dBW arrives at the output of a 1.8 m antenna near −83 dBm by the budget of Table 2: 48 − 205.7 − 0.5 + 44.9 is −113.3 dBW, or the 93.4 dBHz carrier-to-noise density over a −176.9 dBm/Hz floor, and the two agree. Twenty such transponders total about −70 dBm at the LNB input. A commercial LNB with 60 dB of gain and an output compression point near +10 dBm compresses at about −50 dBm of total input and misbehaves ten decibels earlier, which leaves it ten decibels of headroom above the whole satellite load. A terrestrial emitter delivering −50 dBm to the LNB, twenty decibels above everything the satellite sends, drags every carrier down together while a filtered analyzer sees a clean band. The compression figures are illustrative and the LNB's datasheet gives the real ones; the ratio is not, and it is why a filter ahead of the LNB is fitted where the adjacent band has been reallocated to terrestrial service.

The detection-probability argument

A swept analyzer covering span S at resolution bandwidth B listens to any one frequency for the fraction B/S of each sweep. For an interferer of burst length τ and period T_p, the probability of catching it on one sweep is the overlap of the dwell and the burst, and detection over a window follows the geometric law:

Take a rotating maritime radar in the shared Ku uplink segment: twenty revolutions per minute, so a three second period, and a 1.2 degree beam, so ten milliseconds of illumination each turn. Against a 500 MHz survey at 300 kHz resolution sweeping every 150 milliseconds, the dwell in the affected cell is 90 microseconds and the single-sweep probability works out at three and four tenths parts in a thousand. One simplification is made here, and it is generous: the ten milliseconds is treated as a continuous burst, and a radar fills it with microsecond pulses at a kilohertz or so, so the dwell must also coincide with a pulse. The times that follow are floors.

Figure 9 makes the case for engineering the survey instead of scheduling it, under three conditions. The revisit period is the lever. A fast FFT-based sweep crosses 500 MHz in under a millisecond: a real-time analyzer such as the ICX-FieldHawk publishes 577.5 GHz/s at 250 kHz resolution bandwidth with spur rejection standard, so 500 MHz takes 0.87 ms. A sweep that revisits faster than the burst lasts catches every burst that is continuous for its ten milliseconds, so the minutes above belong to the slow sweeper they were worked for. The simplification above returns here with its cost: a radar fills those ten milliseconds with microsecond pulses, and a sweep that passes through the cell eleven times per burst still has to land a pulse inside the short time it spends there, the same coincidence at a faster clock. The fast sweep shortens the wait; it does not remove the coincidence, and only a receiver that holds the cell for the whole burst does. The derivation assumes independence between sweeps, and if the revisit is commensurate with the interferer's period the sweep can alias and never see it. And max hold across many sweeps raises the catch probability exactly as the formula says, but it discards the timing: it records that something happened, never when, and cannot separate two intermittent sources. The fast sweep reduces the first problem and leaves the other two. Figure 10 addresses them.

Figure 10 is the record an intermittent case is built on. The density gives the duty cycle without arithmetic, and a gap-free capture holds the timestamped IQ record of the burst itself, which a swept trace cannot supply.

Sidelobes, and the neighbor two degrees away

The first row of the class table is quantified by a design objective, not a duty cycle. Satellites on the arc are two or three degrees apart, so the station's transmit sidelobes at those angles interfere with the neighbor's uplink. For an antenna of fifty wavelengths or more, which a 1.8 m Ku reflector is at 86, the design objective is a sidelobe envelope of 29 − 25 log₁₀ θ dBi from one degree, or 100λ/D degrees where that is larger (1.17 degrees for the 1.8 m dish at 14.25 GHz), out to twenty degrees off axis. The coordination reference pattern is looser, 32 − 25 log₁₀ θ. Table 6 evaluates both.

Table 6 shows why a two degree spacing requires low sidelobes, and it also shows how pointing error contributes. A 1.8 m beam is 0.8 degrees wide at Ku and, on a Gaussian main-beam model, falls to about 6 dBi at 1.5 degrees, so half a degree of mis-pointing does not put the neighbor in the main lobe. It costs the wanted link about 4.5 dB, and it moves the neighbor from two degrees off axis to one and a half, where the envelope relaxes by only 3 dB, from 21.5 to 24.6 dBi, and where a real pattern's first sidelobe falls. The error is paid twice, once on the wanted carrier and once on the neighbor's. Figure 11 arranges the classes of Table 5 as a decision flow: three questions reach every class.

The requirements, stated so any instrument can be scored

One more number belongs here, because the last row of the decision flow asks for a watch that runs when nobody is on site. A continuous IQ record of a 25 MHz segment runs at 31.25 MSPS, and at four bytes per complex sample, taken as a working figure, that is 125 Mbyte per second, or 10.8 Tbyte a day, more than a station can send to a network operations center. A decision with a ten millisecond IQ snippet attached is 1.25 Mbyte, and a station that sees the radar every three seconds would send twelve hundred an hour if it reported every revolution, or one if it reported the classification. The useful watch raises a decision with its evidence, not a stream, and what counts as a decision is the station's own rule, not a vendor's.

Sections 2 to this one define a criterion, and Table 7 states it as requirements a reader can score any instrument against. None is a sensitivity figure, because sensitivity is not what fails a ground-segment measurement. What fails it is a reading that cannot be reproduced, a timebase worse than the thing it judges, a survey that was too short, or a record without its settings.

10Meeting the requirements: the ICX-FieldHawk and the ICX-400R

An instrument that satisfies Table 7 has to be a real-time analyzer with a published transform relation, a disciplined timebase, coverage through the highest band on the site, and a measurement set on the instrument in the van, not on a price list. The ICX-FieldHawk line is built that way in three enclosures, a handheld, a module presenting over USB or LAN, and the ICX-FieldHawk-R, an IP68 tablet, with the 9.5, 20 and 40 GHz models in each sharing one signal chain. For an earth station carry the tablet: the 40 GHz model, the ICX-400R, covers every band in section 1 and takes the onboard AI option described below, and the 20 GHz ICX-200R covers C band with its harmonic and Ku. Table 8 sets the nine requirements against the published values for the ICX-400R, with the entries that are met only in part stated as such.

The real-time engine, worked

The R1 row of Table 8 is the only one a reader cannot check from a datasheet number alone, so here are the two published relations. The engine transforms the incoming stream in consecutive blocks of N samples at an 8 ns sample period, decimated by D, with no gap between blocks:

At 2048 points with no decimation: 10⁹ / 16,384 = 61,035 frames per second and a probability of intercept of 32.768 microseconds. At 32 points: 10⁹ / 256 = 3,906,250 frames per second and 0.512 microseconds. Both reproduce the published operating points exactly, and that agreement is the check on the relation. A radar burst of ten milliseconds is 610 frames of 16.384 microseconds at the first setting, and the microsecond pulses that fill it, which section 9 set aside, are the second setting's job: at 32 points the engine measures a 0.512 microsecond event at its true amplitude. Nothing in the bottom row of Figure 11, as section 9 modeled it, falls below the second published point; a short-pulse radar mode at a tenth of a microsecond is detected at that setting but not measured at its true amplitude, and either way the survey time of R1 is spent on the emitter's period, as Figure 9 shows, not on the receiver's. The published amplitude accuracy changes at 9.5 GHz, inside Ku, which gives a rule: measure at the L-band IF whatever can be measured there, and at RF only the coupler reading, the regrowth at the amplifier and the harmonic check.

One engine, three ways in

The eight functions of R7 are included, not licensed, and every function is driven three ways. The first is the touch display through the supplied application, SpectraCore. The second is a host over the open API, in C, C++, C#, Python, MATLAB, Qt or LabVIEW, with SCPI as standard, on Windows or Linux. The third, on the ICX-400R or ICX-095R with option 08, is an onboard processor running the station's own code. Figure 12 draws the three as taps on one engine, and the shared engine makes a number taken at the flange comparable with one taken in the rack a year later. For software-defined radio work, the -SDR option, orderable on every current ICX model, adds a SoapySDR driver and a GNU Radio source block that carry the unit's calibration, with every standard measurement function still installed.

The onboard AI module, and what it changes operationally

R9 is the row the third tap in Figure 12 answers, and it changes the maintenance calendar. The ICX-400R and ICX-095R tablets take an optional onboard NVIDIA Jetson Orin NX module, option 08, rated by its maker at up to 117 TOPS (sparse INT8) in the module's highest power mode, which draws up to 40 W. The tablet's published typical power is 10 to 16 W, so the sustained throughput inside the sealed enclosure depends on the power mode the enclosure allows, and that is the condition to carry with the number. The module runs classification and anomaly models the user loads through the open API, on IQ and spectra at the antenna, so the station logs a decision and an IQ snippet instead of a raw stream. No classifier, accuracy, latency or number of signal classes is published for it, so none is claimed here; the claim is architectural, and the arithmetic behind it is in section 9.

11Limitations

• Scalar is not vector, and the tablet has no tracking generator. The ICX-FieldHawk-R datasheet lists none, so the scalar sweep of section 8 uses an external Signal Generator and, for return loss, a bridge or coupler. The handheld's tracking generator options 60, 61 and 62, on the 4.5, 6 and 9 GHz handhelds, reach 9 GHz at most: enough for the L-band inter-facility link and C and X band runs, not for Ku. Scalar gives magnitude only: no phase, no group delay, no vector distance to fault. Waveguide qualification and any measurement headed for a formal report want a vector network analyzer.
• An analyzer does not demodulate the link. The demodulation option measures error vector magnitude on a generic single-carrier modulation at the IF. Energy per bit, error rate and adaptive coding behavior belong to the modem. The analyzer proves the RF chain; the modem proves the service.
• Above 9.5 GHz the amplitude figure is ±3.0 dB, and the input match loosens above 16 GHz. A VSWR of 3.0:1 at a Ka coupler port against a source match of 0.2 adds about ±0.9 dB of mismatch uncertainty on top of the published figure. A 6 dB pad at the input improves the effective return loss by 12 dB; its value goes in the offset.
• The instrument's own phase noise at a Ku carrier is not published. For the ICX-400R the two published points, −107.5 dBc/Hz at 1 GHz and −85.7 dBc/Hz at 40 GHz, both at 10 kHz offset, bracket it; the ICX-200R's −90.6 dBc/Hz at 20 GHz is the nearest published point above Ku. Plan against the conservative end, and judge the reference where the instrument's figure is published and the reference is still what is being read: on a 1 GHz carrier locked to it, where equation (6) adds 40 dB.
• The timebase needs a reference for 0.15 ppm. Published frequency accuracy is under 1 ppm with either oscillator; a 10 kHz offset at Ku is judged with the analyzer locked to an external or GNSS-disciplined 10 MHz reference.
• Sustained AI throughput is a power-mode question. The 117 TOPS (sparse INT8) rating belongs to a module mode drawing up to 40 W, against the tablet's published 10 to 16 W typical. Size a model for the mode the enclosure runs, and measure it.
• Off-site sources can be characterized, not resolved. A field instrument at the victim site establishes an interferer's frequency, bandwidth, timing, polarization and bearing, which opens a case; finding a distant or on-orbit source is multi-site geolocation work with the operator and the regulator.
• You may not need this instrument. A teleport with a carrier monitoring system on every feed and a bench analyzer in the rack already has the weekly and monthly rows of the calendar covered; what it lacks is the flange and the horizon. A C-band-only station that checks its uplink harmonic by other means is covered by the 9.5 GHz ICX-095R, and paying for 40 GHz to measure a 6.4 GHz uplink buys coverage the station does not use. The 20 GHz ICX-200R covers the harmonic at 12.85 GHz and Ku; Ka in the site's future is the reason to pay for 40 GHz.

12Measurement practice, and the workload the instrument is chosen for

The calendar of Table 3 makes one thing plain. Commissioning is one day with the operator on the line and the good reference in the rack. The calendar after it is fifty-two weekly sweeps, twelve mask checks, four sun measurements and one cable sweep a year, every one of them compared with the file, most of them in weather with whatever is in the van. The instrument is therefore chosen for the recurring calendar as much as for commissioning day. The following holds for any analyzer, and most of it needs no additional equipment.

• Set the reference level from the strongest signal at the port, not the one of interest. At the L-band monitor port the strongest carrier may be 30 dB above the one under test, and an attenuator set for the weak one compresses on the strong one with no warning. Confirm with 10 dB of added attenuation: readings that move were compressed.
• Let the instrument warm up for ten minutes before quoting an amplitude, because the published accuracy carries that condition, and enter the reference plane as an offset and record the offset with the trace: coupling factor, pad, test cable, every time.
• Use an RMS detector for power and a peak detector for masks and surveys, and write the resolution bandwidth beside every shoulder reading, because a mask margin changes with it. Survey for the interferer you suspect, not for the time you have: take the period and burst length from the class table, put them through the detection equation, and book that long.
• Never clear a trace without saving it, and never save a trace without its settings. Model, serial, calibration date, firmware, timestamp, weather.

13Summary and selection

Table 9 collects the governing relations and published values with their conditions, so they can be lifted whole into an evaluation matrix.

Table 10 is the selection guide, and the frequency plan decides most of it. A C-band-only station is covered by the 9.5 GHz model, because its uplink stops at 6.425 GHz; checking that uplink's second harmonic near 12.85 GHz, or working Ku, needs the 20 GHz model, and Ka needs the 40 GHz model.

14Definitions, symbols and acronyms

Amplitude offset and reference plane

A constant the analyzer adds to every point of a trace so the display reads power at a chosen plane, the coupler's main line or the feed flange, instead of at its own connector. The offset carries the coupling factor, any pad and the test-cable loss, and it is part of the record.

Antenna factor

The ratio of field strength at a survey antenna to the voltage it delivers, in dB/m. Adding it and the cable loss to a reading in dBµV gives dBµV/m, which is the unit a coordination limit is written in.

Baseline

The dated set of commissioning measurements, with settings and instrument identity, against which every later measurement is compared.

Beam-size correction

The factor by which the sun-method figure of merit is raised to account for the part of the solar disc the main beam does not collect. Near 1 for a small antenna and large for a big one.

Cross-polarization discrimination

The ratio of co-polarized to cross-polarized received field from the same transmission, in dB.

Figure of merit

Receive antenna gain divided by system noise temperature, in decibels per kelvin. The single number that describes a receive station.

Gap-free capture

An acquisition in which every input sample contributes to an output, so no interval exists in which an arriving burst goes unrecorded. A fast sweep still revisits.

Max hold

A trace that keeps the highest value seen in each frequency cell since it was cleared. It records that something happened, never when.

Probability of intercept

The minimum signal duration for which an analyzer measures the amplitude accurately.

Spectral regrowth

Shoulders raised beside a modulated carrier by amplifier nonlinearity, judged against the operator's mask.

Y-factor

The ratio of noise power with the antenna on a hot source to noise power on cold sky, linear or in dB. The measured quantity in the sun method.

15Further reading and one route onward

• ITU-R S.465 and S.580, the sidelobe envelopes of section 9; S.733, G/T determination for earth stations by the sun method.
• ITU-R S.524 and S.728, off-axis radiated power density, with 47 CFR 25.209 and 25.218; P.618 and P.838, the rain attenuation model behind the fade margin.
• Report ITU-R S.2199, broadband wireless access against the fixed-satellite service in 3400 to 4200 MHz, for LNB compression.
• ETSI EN 301 428 and EN 301 443, harmonized standards for Ku and C band terminals; the ICX-FieldHawk-R datasheet for section 10.
• BNC-AN-102, Wide-area spectrum monitoring with automated classification, and BNC-AN-105, RF record and playback, for the unattended watch and the IQ record.

One route onward

If you are writing an acceptance procedure or preparing an interference case, our application engineers would be glad to work the survey duration or the sun measurement for your station.