Frequency Accuracy, Resolution, and the Noise Floor

Your receiver is lying about frequency

A cheap SDR’s reference oscillator is a few parts per million off, and the error scales with frequency. At 100 MHz, 30 ppm is 3 kHz — enough to put an SSB signal in the wrong place and make a band-edge judgement wrong.

Calibrate it once.

1. Tune a signal whose frequency you know exactly. WWV on 5, 10, 15, or 20 MHz is ideal — it is a national frequency standard, so any error you see is yours.
2. Zoom in until you can read the offset. Note how many hertz off it is.
3. Compute the error:

$\text{ppm} = \frac{\Delta f}{f} \times 10^{6}$

A 300 Hz error at 10 MHz is $\frac{300}{10^{7}} \times 10^{6} = 30$ ppm. 4. Enter that value in the software’s PPM correction field. Check a second frequency to confirm it holds — a correct ppm figure fixes the error at every frequency at once, because the error is proportional.

If the correction works at 10 MHz but not at 400 MHz, the oscillator is drifting with temperature. A TCXO-equipped dongle costs a little more and holds calibration; the standard one needs ten minutes to warm up before it settles.

Resolution: what the FFT can and cannot separate

The spectrum display is a Fast Fourier Transform of the incoming samples, and its frequency resolution follows from two settings you control:

$\text{resolution} \approx \frac{\text{sample rate}}{\text{FFT size}}$

At 2.4 MHz sample rate with a 4096-point FFT, each bin is about 586 Hz. Two carriers 200 Hz apart land in the same bin and appear as one signal.

The trade is unavoidable and worth internalising, because it is the same one a spectrum analyser’s resolution-bandwidth control makes:

Try it. Find an FT8 segment — 14.074 MHz on 20 metres — where signals sit about 50 Hz apart. At a small FFT they smear into a band; increase the FFT size and they separate into individual tones. Nothing about the signals changed.

Measuring your own noise floor

Every later observation is relative to this number, so establish it deliberately.

1. Disconnect the antenna. What remains is the receiver’s own noise. Note the level in dBFS.
2. Reconnect it. The floor should rise — typically 6 to 20 dB on HF, less on VHF and UHF.

That rise is the single most useful diagnostic you own:

• Floor rises a lot when the antenna is connected: good. You are hearing atmospheric and man-made noise, which means the antenna is working and the receiver is not the limiting factor.
• Floor barely moves: the antenna, feed line, or connector is at fault. A receiver that cannot hear the sky’s noise cannot hear anything quieter than it either.

On HF a working antenna almost always makes the floor jump, because the atmosphere itself is noisy. On the higher bands the sky is quieter and the rise is smaller — which is exactly why receiver noise figure matters at UHF and barely matters at 40 metres. The measurement module makes that argument; this measurement shows it.

An hour well spent

Do these three things and write the numbers down:

1. PPM correction for your dongle, cold and after ten minutes.
2. Noise floor, antenna connected and disconnected, on one HF and one VHF frequency.
3. The FFT size at which two FT8 signals separate cleanly.

Those numbers characterise your station. Every later measurement — an antenna sweep, a propagation observation, a claim that a band is “dead” — is a comparison against them.

Check yourself

1. Your receiver reads WWV’s 10 MHz carrier at 10.000250 MHz. What PPM correction do you need?
2. You are running 2.048 MHz sample rate with an 8192-point FFT. Can you separate two carriers 300 Hz apart?
3. You connect a new antenna and the noise floor does not move at all. What do you check first?