VNA IF Bandwidth and Averaging: Noise vs Sweep Time

Two knobs, one trade

A vector network analyzer (VNA) has two common ways to clean up a noisy trace: reduce the IF bandwidth, or average several sweeps. Both reduce random noise. Both cost time. This article puts numbers on the exchange so you can choose on purpose instead of by habit.

The IF bandwidth (IFBW) is the width of the receiver filter that sits after the downconversion stage. Noise power collected by the receiver is proportional to that bandwidth. A narrower filter lets in less noise, but it also needs longer to settle and measure each point, so the sweep gets slower.

IF bandwidth: the noise floor moves

Noise power scales with bandwidth, so the change in noise floor between two settings follows directly:

Reducing the IF bandwidth by a factor of 10 lowers the noise floor by 10 dB. Reducing it by 100 lowers it by 20 dB. The sweep time grows by about the same factor: narrowing by 10 times increases it by about 10 times.

This matters most where the signal is small. A stop-band measurement of a filter is the standard case: the true S21 is far below the passband, and the trace sits close to the receiver noise.

On a curve like this, the deep stop-band region is where noise shows up first. In the passband the signal is far above the noise and a wide IF bandwidth is usually enough.

Averaging: the trace gets smoother

Averaging takes N sweeps and combines them point by point. Random noise on the trace falls by:

Averaging 10 sweeps gives a 10 dB improvement. Sixteen averages give about 12 dB. The measurement time is multiplied by about N, so 10 averages take about 10 times as long as one sweep.

Note the shape of the curve. The first few averages help a lot in relative terms, but each further 10 dB costs 10 times more than the last. Going from 10 to 100 averages buys one more 10 dB, and it costs 90 extra sweeps.

Side by side

For random noise, the two options trade at nearly the same rate:

Change	Noise improvement	Time multiplier
IFBW narrower by 10	10 dB	about 10
10 averages	10 dB	about 10
IFBW narrower by 100	20 dB	about 100
16 averages	about 12 dB	about 16
100 averages	20 dB	about 100

So the cost per dB is similar. Pick the option by what else you need, not by a hoped-for speed advantage.

Worked example

Assume a sweep takes 1 s at the starting settings. This is an assumed figure to make the arithmetic concrete, not a measured value. The stop-band trace is noisy, and you want it about 20 dB cleaner.

1. IFBW only. Narrow by 100 times. Noise floor falls 20 dB. Sweep time is about 100 s.
2. Averaging only. Average 100 sweeps. Noise falls 10 log10(100) = 20 dB. Time is about 100 s.
3. Mixed. Narrow by 10 times (10 dB, about 10 s per sweep), then average 10 sweeps (another 10 dB). Total 20 dB, time about 100 s.
4. Partial. If 12 dB is enough, 16 averages give about 12 dB for about 16 s.

All three routes to 20 dB land at about the same time. The sensible move is to ask first how many dB you really need, and stop there.

What neither one fixes

Averaging reduces random noise only. It does not remove drift or systematic errors, and it cannot fix a trace limited by crosstalk. If the stop-band level stops improving as you add averages, the limit is probably not noise. Check connectors, cable movement and calibration before spending more time.

A practical order

1. Start with a wide IF bandwidth and a single sweep to see the whole picture.
2. Work out the improvement you need in dB.
3. Apply it with the IF bandwidth, averaging, or a mix, using 10 dB per factor of 10.
4. Check that the trace actually moved by the expected amount. If not, the cause is not random noise.

In plain words · for students and beginners

IF bandwidth and averaging in plain words

An everyday picture

Imagine listening for a quiet whisper in a noisy room. You can cup your hand around your ear so you hear only a narrow range of sound. That is a narrower IF bandwidth: less noise gets in, but you need to listen longer to catch the whisper. Or you can ask the speaker to repeat the sentence many times and keep what is the same each time. That is averaging: random noises differ every time and fade out, while the whisper stays.

Where it breaks down: if the hum of a fan is always there, repeating the sentence does not remove it. A steady problem is not random, so averaging cannot cancel it.

The same example, in simple steps

A VNA (vector network analyzer) is the instrument that measures how a part passes and reflects a signal. Say one sweep takes 1 s. This is an assumed number for the example. You want a clean trace that is 20 dB lower in noise.

On the dB scale, every 10 dB is ten times the power, so 20 dB is a hundred times less noise power.

1. Each 10 dB of noise reduction costs about 10 times the time.
2. For 20 dB, that is 10 times 10, so about 100 times the time.
3. You can narrow the filter by 100 times, or average 100 sweeps, or do 10 and 10.
4. Either way the sweep takes about 100 s.

Mistakes beginners make

• Averaging too much. The gain per sweep shrinks fast. Going from 10 to 100 averages adds only another 10 dB.
• Expecting averaging to fix everything. It only removes random noise. A drifting setup or leakage between ports (crosstalk) stays in the trace.
• Changing settings after calibration without thinking. Set the IF bandwidth and number of averages you plan to use, and keep them for the measurement.
• Using the narrowest setting for every trace. A strong passband signal does not need it. You pay time for nothing.

Remember

• 10 times narrower IF bandwidth gives 10 dB less noise and takes about 10 times longer.
• N averages give 10 log10(N) dB, so 16 averages give about 12 dB.
• Neither one fixes drift, systematic errors or crosstalk.

Related pages on RF View

• What S21 and S11 of a Low-Pass Filter Look Like