A Practical Guide to Real-Time Spectrum Analysis

Real-time spectrum analysis takes a different stance. Instead of visiting frequencies one at a time, it captures a wide block of spectrum all at once and keeps capturing it without pause. The instrument transforms a continuous stream of samples into a continuous stream of spectra, so a burst that lasts only microseconds still lands inside a measurement window.

This matters more every year. The spectrum is crowded. Frequency-hopping radios change channels thousands of times a second, radar pulses come and go in microseconds, and interference often shows up as a short spit of energy that never appears twice in the same place. Agile, intermittent, and overlapping signals are common. A swept analyzer can spend its whole sweep looking at the wrong frequency at the wrong instant. A real-time analyzer is built so that does not happen.

Swept / superheterodyne

The classic design. A local oscillator tunes the receiver across the span while a narrow filter measures power at each step. The strength is reach and resolution: you can sweep enormous spans and resolve very close tones with a narrow resolution bandwidth. The trade is time. The analyzer only ever looks at one slice at a time, so anything happening elsewhere during that slice is invisible. For stable carriers and slow drift it remains an excellent, economical tool.

Vector signal analyzer (VSA)

A VSA digitizes a slice of spectrum and preserves both magnitude and phase, which lets it demodulate. It will hand you constellation diagrams, error vector magnitude, and the inner structure of a modulated signal. The strength is depth of insight into a known signal. The trade is that a VSA is built to dissect a signal you already expect, not to stand watch over a whole band waiting for something unexpected to flash by.

Real-time (RTSA)

An RTSA also digitizes a wide slice, but it processes that slice continuously and without gaps. Its strength is certainty of capture: if a signal occurs inside the analysis bandwidth, the instrument is guaranteed to see it down to a known minimum duration. The trade is that real-time bandwidth is finite, so you watch a defined window rather than the entire span at full speed. Many modern analyzers blend the three, sweeping for wide coverage and switching to real-time inside the band that demands it.

First, the analyzer digitizes a wide block of spectrum. A downconverter places the band of interest at an intermediate frequency, and a fast analog-to-digital converter samples it as in-phase and quadrature data. The width of this captured block is the real-time bandwidth, and it sets the largest window the instrument can watch all at once.

Second, the analyzer runs a Fast Fourier Transform on that sample stream to turn time-domain samples into a spectrum. What makes the analysis real-time is that the FFTs never stop: the engine transforms one block, then the next, with no dead interval between them. Some designs also overlap successive windows; the ICX-FieldHawk engine does not, and runs its frames back to back, strictly gapless and overlap-free. An event that straddles a frame boundary is split between two frames, which is why the guaranteed capture duration is two frames rather than one (section 4).

Third, the result is gap-free coverage. Spectra arrive in a contiguous train, each one a fresh photograph of the whole real-time band. Stack enough of them per second and the picture is effectively a movie of the spectrum with no missing frames. That contiguity is the entire point, and it is what the next section quantifies.

Two things drive POI. The first is the FFT rate, meaning how many transforms the engine completes each second. The faster the spectra arrive, the shorter the event that still lands inside one of them. The second is the frame length. For a gapless engine the guaranteed duration is two frames: a burst that long falls entirely inside at least one frame whatever its timing, so it is seen at full amplitude. The ICX-FieldHawk datasheets give both as arithmetic, frame rate = 10⁹ / (N × D × 8 ns) and POI = 2 × N × D × 8 ns, where N is the FFT size and D the decimation factor. A 32-point transform at decimation 1 gives 3,906,250 frames per second and a POI of 0.512 µs; 2,048 points give 61,035 frames per second and 32.768 µs. A shorter transform shortens the POI and widens the resolution bandwidth, which is the trade the next section is about.

One distinction is worth stating plainly. There is a difference between POI for detection and POI for amplitude accuracy. Detection POI is the shortest event the analyzer will reliably notice at all, often quoted at a modest level below the true amplitude. Amplitude POI is the longer minimum needed for the measured level to be within a tight tolerance, frequently within a fraction of a decibel, of the real value. A very small POI figure can be correct for detection alone, so read which one is quoted. Detecting a 5 microsecond pulse is one claim. Measuring its power within half a decibel is a stronger one.

Resolution bandwidth (RBW) is how finely the analyzer can separate two close tones. A narrow RBW resolves detail but takes longer to compute and reduces how fast spectra can arrive. Span is how wide a frequency range you display. A wide span shows more of the world at once but, for a given FFT size, spreads the resolution thinner. The link between them is the FFT itself: the number of points and the captured bandwidth together fix the bin spacing, and the bin spacing sets the achievable RBW.

Windowing is the third lever. Before each FFT, the sample block is multiplied by a window function that tapers its edges to zero. Without it, the abrupt block boundaries smear energy across the spectrum, an effect called spectral leakage, and small signals near large ones disappear into the skirts. The choice of window trades main-lobe width against side-lobe height. A flat-top window gives excellent amplitude accuracy at the cost of frequency resolution, while a window like Hann or Blackman favors separating nearby tones. There is no universally correct window. There is only the right window for what you are trying to see.

Instead of plotting a single trace, the analyzer accumulates many spectra into a two-dimensional histogram. The horizontal axis is frequency, the vertical axis is amplitude, and color encodes how often each point in that grid was hit. A signal that is present nearly all the time burns in as a hot, saturated color. A signal that flashes for a microsecond once in a while paints a faint, cool wash. The display turns frequency of occurrence into something you can read at a glance.

The payoff is separation. A weak, intermittent emitter sitting underneath a strong, steady carrier is nearly impossible to see on a single trace. On a density display the two paint in different colors because they occur with different frequencies, and the eye picks them apart immediately. Bright means persistent, dim means rare, and structure in the color tells you the duty cycle of everything sharing that band.

On the ICX-FieldHawk the capture is armed in IQ Streaming mode, which acquires on multiple trigger methods, and through the external trigger input, which responds up to 500 times per second. Each trigger starts an IQ capture of up to 100 MHz bandwidth into 128 Mbyte of built-in memory, and continuous recording runs at up to 25 MHz. A frequency-domain condition can be applied on the host: the API delivers each spectrum, and the host’s own code decides when to start a capture. Either way the recorded IQ, not the live display, is what the analysis is done on.

A wide real-time bandwidth with a slow FFT rate still leaves brief events at risk. A blazing FFT rate over a narrow band cannot watch much of the spectrum at once. Strong sensitivity is wasted if the dynamic range lets a loud signal swamp the quiet one beside it. Judge the set as a whole, against the signals you actually expect to chase.

The FFT engine runs in the FPGA, and the host software, SpectraCore, presents the real-time density display, the spectrogram and the historical trace, with IQ Streaming mode acquiring on multiple trigger methods into 128 Mbyte of capture memory. The same engine runs across three form factors so the measurement does not change when the deployment does. The handheld unit is built for walk-and-find work, the IP68 ICX-FieldHawk-R is built for field and vehicle environments, and the USB/LAN ICX-Series turns a laptop or an embedded host into a full real-time analyzer for benches, racks, and remote sensors. Pair any of them with the RFA-3100-N directional antenna to turn detection into a bearing.

To talk through real-time bandwidth, POI, and the right form factor for your mission, contact Berkeley Nucleonics at 800-234-7858 or info@berkeleynucleonics.com.

For a quick question, chat with an engineer at berkeleynucleonics.com.