Spectrum analyzer basics · Guide

Spectrum analyzer settings explained

A spectrum analyzer has a lot of knobs, but most of them answer one of three questions: which part of the spectrum am I looking at, how much detail do I want, and how strong are the signals? This guide explains each setting in plain words, first for bench and radio analyzers, then for the FFT analyzers used for sound. Play with the box below as you read: it shows the trade between detail and speed live.

Try it

The test sound has two tones only 10 Hz apart (1,000 and 1,010 Hz), a very faint tone at 1,200 Hz and a short beep at 1,500 Hz. Change the FFT size and the window and watch what you can see.

FFT size

Window

Press Play or pick a setting to start

Bin width —Frame —1,000 + 1,010 Hz —Faint 1,200 Hz —Beep —

Centre, span, start and stop

These four settings choose the slice of the spectrum on the screen. Start and stop are the left and right edges. Centre is the middle, and span is the width from edge to edge. They are two ways to say the same thing: a centre of 100 MHz with a span of 20 MHz is the same view as start 90 MHz and stop 110 MHz.

Use centre and span when you know roughly where a signal is and want to zoom in on it. Halve the span and you see twice the detail around the same spot. Use start and stop when you care about a band with fixed edges, like the 2.400 to 2.4835 GHz Wi-Fi band or 20 Hz to 20 kHz for sound.

Span also changes other settings. On most bench analyzers the RBW follows the span when it is set to "auto", so a narrow span gets a narrow filter. On an audio analyzer the span is just the visible range: 20 Hz to 1 kHz is a good view for hum, and 20 Hz to 20 kHz on a log scale is the normal view for music.

RBW and VBW

RBW means resolution bandwidth. A swept analyzer moves a narrow filter across the span and measures how much power gets through at each step. RBW is the width of that filter. It decides three things at once.

  • What you can tell apart. Two signals closer together than about one RBW merge into one bump. To see two carriers 5 kHz apart, you need an RBW of a few kHz or less.
  • How low the noise floor sits. Noise power grows with filter width, so every 10 times narrower RBW lowers the displayed noise by 10 dB. A weak signal hidden at 100 kHz RBW can pop out at 1 kHz RBW, 20 dB lower.
  • How slow the sweep is. A narrow filter needs time to settle, and the sweep time grows with the span divided by the RBW squared. Ten times narrower RBW can mean 100 times slower sweeps.

RBW usually comes in 1-3-10 steps, for example 1 kHz, 3 kHz, 10 kHz. Leave it on auto while you look around, then narrow it when you need to separate close signals or see a weak one. On auto, many analyzers keep the RBW at roughly 1 % of the span, so zooming in narrows it for you. To work out how far the noise floor drops for a given RBW, use the noise floor and RBW calculator.

VBW means video bandwidth. It is a low-pass filter applied after the signal has been detected, so it smooths the trace instead of sharpening it. A narrow VBW turns a fuzzy, jumping noise floor into a steady line, which makes a small signal just above the noise easier to read. It does not change which signals can be told apart; only RBW does that.

A common rule is to set VBW to RBW ÷ 10 when you hunt for weak signals and VBW equal to RBW (or 3 times it) when you measure fast-changing or pulsed signals, which a narrow VBW would flatten. Like RBW, a narrow VBW slows the sweep.

Reference level and attenuation

The reference level is the level of the top line of the screen, for example 0 dBm. Everything is drawn below it, usually 10 dB per division over 10 divisions, so 0 dBm at the top puts the bottom line at −100 dBm. Set it just above your strongest signal. Too high, and small signals sit in the noise at the bottom. Too low, and the strong signal runs off the top and may overload the input.

Input attenuation is a switchable pad in front of the analyzer's first mixer. It protects the input and keeps strong signals in the range where the analyzer is linear. The price is noise: every 10 dB of attenuation lifts the noise floor by 10 dB. On most analyzers the attenuation follows the reference level unless you set it by hand.

Watch for overload. If extra peaks appear that should not be there, add 10 dB of attenuation. A real signal stays at the same level on the screen, because the analyzer corrects its reading for the pad; only the noise floor rises. Distortion made inside the analyzer drops or vanishes. That simple test tells you whether a spur is real or home-made. Never feed more than the maximum input level printed next to the connector; a few watts from a radio can destroy the front end.

Detectors and trace modes

Each pixel across the screen covers a small range of frequencies, and often many measurements land in that one pixel. The detector decides which value is drawn.

  • Peak shows the highest value in each pixel. It never misses a short or narrow signal, so it is the default for finding carriers and spurs. It makes noise look higher than it is.
  • Sample shows one value from each pixel. It gives an honest picture of noise, but can miss a narrow signal that falls between samples on a wide span.
  • RMS (power average) shows the true average power in each pixel. Use it for noise, for digital signals like Wi-Fi, LTE or DAB that look like noise, and for channel power.

Trace modes decide what happens between sweeps. Clear/write draws each sweep fresh. Max hold keeps the highest value seen at each frequency, which is perfect for catching a signal that hops or comes and goes. Min hold keeps the lowest. Average mixes several sweeps so the noise settles and steady signals stand out. Many analyzers show two or three traces at once, so you can keep a max hold trace next to a live one.

FFT size and windows on audio analyzers

Audio analyzers, including the live analyzer on this site, do not sweep. They take a block of samples and run an FFT, which measures every frequency at once. The settings have different names, but the trade-offs are the same.

FFT size is the number of samples in one block. The frequency step, called the bin width, is the sample rate divided by the FFT size. At 48 kHz, 8,192 samples give 5.86 Hz bins and a frame of 171 ms; 32,768 samples give 1.46 Hz bins but a frame of 683 ms. The bin width plays the part of RBW: smaller bins separate closer tones and lower the noise floor. The frame length plays the part of sweep time: a bigger FFT reacts more slowly and smears short sounds. In the Try it box, the two tones 10 Hz apart only split into two peaks from 16,384 points, and the 150 ms beep only blinks clearly at 4,096 points and below.

The window shapes each block before the FFT so that sound cut off at the block edges does not spread across the spectrum. Rectangular (no window) gives the sharpest peaks but spills energy far to the sides, which hides the faint tone in the Try it box. Hann is the everyday choice. Blackman-Harris pushes the spill down further for very quiet details next to loud ones. Flat-top reads the level of a tone most exactly, at the cost of wider peaks. The page on window functions compares them side by side.

Try this order in the box above. Start at 8,192 with Hann: the faint tone at 1,200 Hz shows, 80 dB below full scale, but the two close tones are one blob. Go to 16,384 and they split, while the beep turns into a steady smear. Now pick Rectangular: the faint tone drowns in the spill from the loud pair. Pick Blackman-Harris and drop to 2,048: the faint tone comes back even with coarse bins, and the beep blinks again. No single setting shows everything, which is why good analyzers let you change them.

Averaging and smoothing calm a jumpy display, like VBW does. Exponential averaging follows changes smoothly; linear averaging over a set number of frames gives a steady result for noise measurements. Octave smoothing (1/3 or 1/6 octave) blends nearby bins and makes music and room curves easier to read. Work out bin width and frame time for any sample rate with the FFT resolution calculator.

Settings cheat sheet

Good starting points by job. Change one setting at a time from here.

JobSettings to start with
Find mains humFFT 32,768, Hann, 20 Hz to 1 kHz on a linear scale, averaging 4 to 8 frames
Tune an instrument or find a noteFFT 8,192 to 16,384, Hann, no averaging, log scale 50 Hz to 5 kHz
Look at a mixFFT 8,192, Hann, slow exponential averaging, tilt 4.5 dB per octave, 1/6-octave smoothing
Measure a tone's level exactlyFlat-top window, FFT 8,192 or more, peak readout
Check a room or speakerPink noise, 1/3-octave RTA or 1/6-octave smoothing, slow averaging
Find a weak RF carrierNarrow span around it, RBW 1 kHz or less, VBW = RBW ÷ 10, reference level just above the strongest signal, low attenuation
Catch a signal that comes and goesPeak detector, max hold trace, span over the whole band
Measure noise or a digital channelRMS detector, trace average, RBW well under the channel width

New to the screen itself? Start with what is a spectrum analyzer, or browse more lessons in spectrum analyzer basics.

FAQ

What is RBW?

The width of the analyzer's measuring filter. Narrow RBW shows close signals apart and lowers noise, but sweeps slower.

What is VBW?

A smoothing filter after detection that makes the trace less noisy without changing the resolution.

What reference level should I use?

Set it just above the strongest signal so it fills the screen without overloading the input.

What is the audio equivalent of RBW?

The FFT bin width: sample rate ÷ FFT size. At 48 kHz and 8,192 points it is 5.86 Hz.

Written and tested by the SpectrumAnalyzer.app team · Last checked

How we tested: Checked with the test sound in the Try it box: two tones at 1,000 and 1,010 Hz show as two peaks from an FFT size of 16,384 at 48 kHz (2.93 Hz bins) and as one blob at 2,048, the tone 66 dB down at 1,200 Hz is hidden by the Rectangular window and visible with Hann, and the 150 ms beep is only sharp with frames of 150 ms or less.