Averaging
Understanding Averaging in RF Measurements
The displayed noise on an RF instrument is a combination of the instrument's internal noise (thermal noise in the front-end amplifier, quantization noise in the ADC) and the noise content of the signal itself. On a single sweep, this noise appears as random fluctuations superimposed on the true signal level. By combining multiple sweeps, the random component averages toward zero while the deterministic signal component remains unchanged. The result is a cleaner, more repeatable trace with improved ability to resolve low-level signals.
Averaging is not free. Each additional average requires another complete sweep, increasing total measurement time linearly. A 100-sweep average on a spectrum analyzer with a 1-second sweep time takes 100 seconds. In production test environments where throughput matters, the engineer must balance noise reduction against measurement speed. VNA measurements face the same trade-off: increasing the IF bandwidth reduces noise faster than averaging but sacrifices frequency selectivity.
Noise Floor Improvement
NFreduction = 10 · log10(N) dB
Examples:
N = 10 averages: 10 dB improvement
N = 100 averages: 20 dB improvement
N = 1000 averages: 30 dB improvement
Log-Power Averaging Bias (for noise-like signals):
Bias = −2.51 dB (systematic underestimate of true noise power)
Use RMS/power averaging mode for accurate noise power measurements. Log-power averaging is only valid for CW signal level measurements.
Averaging Types Compared
| Type | Where Applied | How It Works | Noise Reduction | Best For |
|---|---|---|---|---|
| Trace Averaging | After detection | Combines N complete sweeps point-by-point | 10 log(N) dB | Spectrum analyzer: revealing weak signals |
| Video Averaging | Video filter | Lowpass filter on detected signal within one sweep | Less effective (correlated samples) | Quick smoothing of displayed trace |
| Sweep Averaging | After detection | RMS combination of N sweeps | 10 log(N) dB | VNA: improving dynamic range |
| IF Bandwidth Reduction | Before detection | Narrows IF filter, increasing sweep time | 10 dB per decade of BW reduction | Spectrum analyzer: maximum sensitivity |
Frequently Asked Questions
How much does averaging improve the noise floor of a spectrum analyzer?
The displayed noise floor drops by 10 × log10(N) dB, where N is the number of averages. Ten averages give 10 dB, 100 averages give 20 dB. This improvement applies only to random noise. Coherent signals like CW tones and spurs remain at their true level because they repeat identically on every sweep. This selective reduction is what makes averaging useful: it separates signals from noise.
What is the difference between trace averaging and video averaging?
Trace averaging combines complete sweep results mathematically, with each frequency bin averaged across N independent sweeps. Video averaging applies a lowpass filter to the detected signal within a single sweep, smoothing the trace in real time. Trace averaging is more effective per unit time because each sweep provides statistically independent noise samples. Video averaging is quicker to set up but yields less noise reduction because successive samples within one sweep are correlated through the RBW filter.
Does averaging affect the accuracy of signal level measurements?
For CW signals, averaging does not change the measured amplitude because the signal is deterministic. For noise-like signals (OFDM, CDMA, digitally modulated carriers), log-power averaging underestimates true power by 2.51 dB. This occurs because the logarithm function is nonlinear: the average of the log is not the log of the average. Switching to RMS or power averaging mode eliminates this bias and provides correct noise power readings.