Signal Explorer
Signal Explorer beta
This is a beta. It is new and still being worked on, so you may find rough edges, and it is not guaranteed to behave on every browser or device. If something looks wrong or does not work, please tell me or leave a comment at the bottom of the page. That feedback is exactly what I need at this stage.
A radio simulator you can hear. Pick a mode, push it around with the controls, and watch every view change together. Turn the sound on to hear the audio going in and what comes back out of the receiver.
Arrived from Seeing signals: I and Q from first principles? Every panel of Figures 4, 5 and 6 is here, marked with its number. Each one has its own link, so #5d opens the airband panel directly.
The view
Slowing is not a display setting. It changes how much signal passes in a frame and nothing else, so every frequency, deviation and symbol rate stays exactly as stated and a measurement taken at any speed is still valid. It is also why the sound switches off below a quarter speed: the loudspeaker is playing these same samples, and there is no sound left in a signal slowed five hundred times.
Trace colours the last few thousand samples from purple through blue and green to yellow, so you can see which way the point is going. Density piles every sample into a grid and colours by how many landed in each square, which is how the printed figures were made. Real signals move far too fast to watch, and that is why.
Each mode starts with a trail about one revolution long, which is what makes the colour run once round the shape. Lengthen it a long way and every colour is painted over by the newest one, so the trail turns a single flat green and the direction of rotation disappears.
How coarse the density map is. This is not just cosmetic: the printed figures vary by a factor of seven in it, and a coarse grid turns a smooth gradient into what look like distinct contour rings. Several of the printed panels are banded for exactly that reason.
Sound
Your browser will not start audio without a click, which is why this is a button rather than a switch.
The second is the whole signal chain: modulated, put through the channel, and detected. Turn the noise up, mistune it, or push AM past 100 per cent, and the difference between the two is the difference the channel made. Both are the same samples: what the plots draw is what the loudspeaker plays, so the two can never disagree.
The channel
The rotation rate is the tuning error, exactly. The readout below reports the rotation of the signal itself, so it will agree with this slider whatever the speed control is set to.
Noise is the receiver's own, so it is added after the gain. Turning the gain up lifts the signal out of the cloud; it does not shrink the cloud.
A sum of sinusoids approximation to a multipath path, not a measured channel. It is here because the airband panel is the same modulation as the broadcast one and looks nothing like it.
The signal
The index is the audio swing as a percentage of the carrier. Past 100 per cent the envelope goes negative, the point passes straight through the origin, and the carrier is overmodulated. Worth doing once, and worth listening to.
Loaded in the browser and never sent anywhere. Speech on SSB is the one to try: watch the cloud breathe.
Measured, this instant
Taken from the same samples the plots are drawing, not from the settings. If a number here disagrees with a slider, the number is right.
Where the numbers come from
Each mode carries the sample rate, tuning and filtering of the GNU Radio flowgraph that produced the matching capture in the article, so the shapes here are generated under the same conditions the printed ones were recorded under. Parameters the flowgraphs do not record, such as the sideband of an SSB capture or the tone frequencies behind the two tone rosette, are marked as standard values rather than quietly invented. The origin line under each mode says which is which.
One thing here is not a reproduction. The published figures are densities accumulated over 100000 samples, roughly four tenths of a second, because a constellation at its real speed is a blur. The speed control exists so the same signal can be slowed to something a person can follow. Slowing it changes how much signal passes per frame and changes nothing about the signal, so every frequency, deviation and measurement stays exactly as stated.
Comments
You don't need an account, just a name. Comments are read before they appear, so there may be a short delay before yours shows up.