What this lab shows
An IQ recording keeps a slice of the radio spectrum rather than one station: everything the receiver could hear across its sample rate went into the file at once. This page draws that slice as a waterfall, frequency across and time downwards, and puts a software receiver on top of it, so the stations on a recording made last week can be tuned in one by one today.
The question it explores: what does a band look like from the inside, and what does a receiver do to pull one signal out of it?
Things to try
- Press Try the test recording, then Play: the AM station at 7.1300 MHz plays a tune. Click the narrow FM station at 7.0400 MHz, choose NFM and compare the two.
- Choose the RTTY signal at 7.0750 MHz in the list under Receiver: the receiver goes to 7.0735 MHz in USB, 1.5 kHz below the signal, so its two tones come out at 1415 and 1585 Hz and you hear the warble. Save 0 to 20 s as WAV and open the file in the RTTY Decoder: the text comes out.
- Tune the USB station at 7.1750 MHz, then switch the mode to LSB: the tune is gone, because LSB listens below the dial and there is nothing there. Type 7.178 as the frequency: in LSB the tune comes back upside down, with the high notes lower than the low ones.
- Tune the Morse signal at 7.0000 MHz in CW and you hear a 700 Hz tone. Untick “Width matches the mode” and widen the filter from 500 Hz to 2 kHz: the hiss around the tone grows, while the tone stays the same.
- Zoom in on the narrow FM station with the + buttons for the band and for time until single notes show. Each note is a comb of lines spaced by its own pitch, wider apart for higher notes: the frequency swings hundreds of times a second, far faster than a row of the waterfall, so you see sidebands rather than a moving line.
What to look for
- A carrier is a thin steady line. AM adds faint sidebands on both sides of it, CW is the same line broken into dots and dashes, and FM is a band whose width changes with the sound. SSB has no line at all, only energy on one side of where the carrier would be. RTTY is two lines 170 Hz apart that take turns.
- Real recordings from 8-bit receivers often have a spike exactly at the centre. It is the receiver’s own DC offset, not a station. Faint mirror images of strong signals on the other side of the centre come from I and Q not being perfectly balanced.
- “In the filter” compares the strongest point inside the receiver’s filter with the median of the visible band: a quick measure of how far a signal stands above the noise. For a careful reading use the two cursors.
Supported recordings
- WAV with two channels, I on the left and Q on the right: 8, 16, 24 or 32-bit integers, or 32 or 64-bit floats, also in the RF64 form used for files over 4 GB. SDR#, SDR++ and HDSDR save their baseband (IQ) recordings this way. The centre frequency comes from an auxi chunk when the file has one, or from the file name.
- Raw IQ without a header: cu8 (unsigned 8 bit, as written by rtl_sdr and rtl_433), cs8 and cs16 (signed 8 and 16 bit), and cf32 (32-bit float, also the .cfile of GNU Radio and the .raw recordings of gqrx). The page takes the format from the extension and the frequency and sample rate from the name where it can, for example
gqrx_20261114_120000_100100000_2400000_fc.raw,…_100.1MHz_2.4Msps.cu8or…_433.92M_250k.cu8; otherwise you enter them. - SigMF: a .sigmf-meta and .sigmf-data pair (choose both files at once) or a .sigmf archive, with complex samples of 8, 16 or 32-bit integers or 32 or 64-bit floats in either byte order. The sample rate, centre frequency and start time come from the metadata.
Getting a recording
- In your SDR program, record the baseband (some programs call it the IQ or RF recording), not the audio. An audio recording has one channel; it already holds what the loudspeaker played and goes straight into a decoder such as the Signal Identifier.
- From the command line,
rtl_sdr -f 100.1M -s 2.4M -n 24000000 broadcast_100.1MHz_2.4Msps.cu8records 10 seconds. With the frequency and rate in the name, the page opens it without asking. - Mind the size: at 2.4 MS/s an 8-bit recording grows by about 290 MB a minute, and 16-bit samples take twice that. The page reads only the parts it needs, so a file far larger than the memory of the device still opens.
How it works
IQ samples
The receiver shifts the band down to zero and samples two signals a quarter of a cycle apart, I (in phase) and Q (quadrature). Together they make one complex number per sample, which tells frequencies above the centre from those below it, so a sample rate fs covers a band fs wide around the centre frequency fc:
fc − fs/2 to fc + fs/2
Waterfall
Each row is a fast Fourier transform of N samples with a Blackman-Harris window, and bin k lies at
f = fc + (k − N/2) · fs / N
Levels are in dBFS, where a full-scale tone reads 0 dBFS. When a column of the picture covers several bins it shows the strongest, so narrow signals stay visible when the whole band is on screen, and a row of the overview averages up to eight transforms spread over its stretch of time. Zooming in uses longer transforms, up to 65,536 points, which sharpens the frequency at the cost of time: a bin fs/N wide needs N/fs seconds of signal.
Receiver
- Mixer: multiplying the samples by e−j2πΔf·n/fs moves the dial frequency, Δf away from the centre, to 0 Hz. In CW the mixer sits 700 Hz lower, so a carrier on the dial comes out as a 700 Hz tone. Moving the wanted signal to one fixed frequency before filtering it is the superheterodyne principle, for which Lucien Lévy and Edwin Armstrong both filed patent applications in 1917; here the fixed frequency is zero.
- Half-band filters halve the sample rate step by step, down to 24 to 48 kHz (200 to 400 kHz for broadcast FM). Every other coefficient of a half-band filter is zero, so each halving costs little.
- Channel filter: 1,025 taps (129 for broadcast FM) of the chosen width, centred on the dial for AM, FM and CW, above it for USB and below it for LSB.
- Detector: for AM the size of the filtered signal |b|; for FM its frequency, f = Δφ · fIF / 2π, from the phase step Δφ between two samples; for SSB and CW the real part of b, which is what a product detector gives.
- Broadcast FM: a de-emphasis of 50 or 75 µs undoes the treble boost the transmitter adds.
- Automatic gain control for AM, SSB and CW, then a windowed-sinc resampler to the rate of the sound output (48 kHz in a saved WAV).
In AM and CW a click or tap lands on the nearest carrier within about 12 pixels: the page picks the strongest line within one filter width in the waterfall, so a sideband never wins over its carrier, and then refines it with a fine spectrum (four 4096-point transforms, about half a second of signal) at the low rate that the half-band filters of step 2 give. That matters for CW, where a 500 Hz filter set a pixel off would miss the carrier altogether.
Live and computed sound
Live playback reads the recording in pieces of 60 ms and keeps about half a second of sound queued; a new tuning starts again from what you are hearing. To keep up, the page has to process the whole sample rate in real time, 2.4 million samples a second for a common 8-bit receiver. “Compute, then play” works out the stretch first instead.
Limits
- Broadcast FM is mono: no stereo and no RDS.
- No squelch, noise blanker or notch filter, and no digital modes: save the sound as WAV and open it in a decoder.
- Live playback needs a device that can process the full sample rate. At 10 MS/s or on a slow phone the sound may break up; the page then says so.
- Without a centre frequency the scale shows offsets from the centre in kilohertz.
- SigMF annotations and recordings with several channels are not read; when a recording has several captures, the frequency of the first is used.
- The test recording is made up: its stations sit at round offsets and do not follow a real band plan.
Controls
- Buttons: Open IQ file (or drop the file on the display), Try the test recording, Play or Stop; the modes AM, NFM, WFM, USB, LSB and CW; − step and + step; zoom of the band and of time, Centre on tuning, Whole recording; Auto levels; what a click on the display sets (Tuning, Cursor A, Cursor B) and Clear cursors; Save the picture, Save as WAV, Compute, then play.
- Sliders: volume (a limiter keeps the output about 6 dB below full scale), position in the recording, filter width, gain when AGC is off, and the floor and range of the colours.
- Mouse: click a signal to tune to it; drag to move along the band and through time; Ctrl with the wheel, or a pinch on a touchpad, zooms the band around the pointer; a click on the time scale at the left jumps there.
- Touch: tap a signal to tune to it, slide sideways to move along the band, pinch with two fingers to zoom it; a vertical swipe scrolls the page.
- Keyboard: with the display focused, Space plays or stops, ← and → tune by one step (with Shift by ten), ↑ and ↓ move 1 s back or forward (with Shift 10 s), + and − zoom the band, [ and ] zoom time, 0 shows the whole recording and M changes the mode. Esc stops the sound anywhere on the page.
Frequently asked questions
Why does my WAV file open as sound and not as IQ?
Because it has one channel: the program saved the demodulated audio, not the baseband. Such a file goes straight into the Signal Identifier or a decoder. For an IQ recording, record again with the baseband option.
I can see a signal but hear only noise. What is wrong?
Usually the mode or the dial. SSB played in AM is a mumble, a USB signal in LSB is silent, and a CW carrier outside a 500 Hz filter cannot be heard at all. Zoom in until the dashed lines of the filter cover the signal, and match the mode to its shape on the waterfall (see What to look for).
Sources
- The SigMF project (accessed 14 November 2026). Signal Metadata Format. The datatypes such as cf32_le and ci16_le, the fields core:sample_rate, core:frequency, core:datetime and core:sample_start, and the archive form.
- Wikipedia contributors (accessed 14 November 2026). FM broadcasting, section on pre-emphasis and de-emphasis: 75 µs in the Americas and South Korea, 50 µs elsewhere.
- Wikipedia contributors (accessed 14 November 2026). Superheterodyne receiver. The patent applications of Lucien Lévy and Edwin Armstrong in 1917, and the principle: convert the received signal to a fixed intermediate frequency and filter it there.