VNA Dynamic Range and Noise Floor in Stop-Band Measurements

Transmission dynamic range is the source power minus the receiver noise floor, in dB. Raising the source power by 10 dB or narrowing the IF bandwidth by a factor of 10 each gains 10 dB, until the receiver compresses.

For practitioners

What dynamic range means for S21

A VNA measures transmission by comparing the power at the receiver with the power sent by the source. The ratio, in dB, is S21:

S21 (dB)=10log⁡10PreceivedPsourceS_{21}\,(\text{dB}) = 10\log_{10}\frac{P_{\text{received}}}{P_{\text{source}}}

The receiver can only report a signal that rises above its own noise. So there is a smallest power it can resolve, the noise floor. The transmission dynamic range is the distance between the largest signal you can send and that floor:

DR (dB)=Psource (dBm)−Pfloor (dBm)DR\,(\text{dB}) = P_{\text{source}}\,(\text{dBm}) - P_{\text{floor}}\,(\text{dBm})

This is the key figure for stop-band work. A filter that rejects 100 dB cannot be measured on an instrument with 90 dB of dynamic range, however good the calibration is.

A transmission level of −90 dB means the received power is 10^-9 of the source power. That is a very small ratio, and a receiver has to separate it from its own noise.

+30 dBm1 W+20 dBm100 mW+10 dBm10 mW0 dBm1 mW-10 dBm100 µW-20 dBm10 µW-30 dBm1 µW+10 dB = 10 times the power-3 dB = about halfdBm is power compared with 1 mW. 0 dBm is one milliwatt.
Power levels on the dBm scale. Every +10 dB is ten times the power.

Why the stop-band flattens at the floor

Follow the S21 trace of a low-pass filter from the pass-band into the stop-band. Rejection grows, and the trace falls. When the true signal reaching the receiver drops to the noise floor, the trace stops falling. From there it shows receiver noise, not the device.

The result is a noisy, flat curve. It is not a true measurement of the device's rejection. The filter may reject much more than the trace shows. The flat part tells you the floor of the instrument, nothing more.

0-10-20-30-4000.511.522.53cutoff 1 GHz: -3 dBS21 (what passes)S11 (what bounces back)Frequency (GHz)dB
S21 and S11 of the low-pass sample filter, computed from the circuit model. Not measured data.

The figure above is computed from a circuit model, not measured. A model has no noise, so its stop-band keeps falling. A real trace on a VNA stops at the floor.

A quick check: if the flat region sits at the same level for a very different device, you are looking at the instrument. If the level moves when you change the IF bandwidth, you are looking at noise.

Lever 1: raise the source power

Raising the source power by 10 dB raises the measured signal by 10 dB. The noise floor of the receiver does not move. So the dynamic range improves by 10 dB, and the stop-band trace has 10 dB more room before it meets the floor.

There is a limit. The receiver has a maximum level it handles linearly. Past that point it starts to compress, and the pass-band readings become wrong. The gain in dynamic range holds only until the receiver starts to compress.

Two practical points follow:

  • The compression limit is set by the strongest signal at the receiver, which is the pass-band level. A device with gain, or a low-loss pass-band, puts more power on the receiver.
  • The stop-band signal is small, so it is the pass-band that decides how far you can raise the source.

If the receiver needs protecting, or the device cannot take the power, reduce the level at the right place. Do not just lower the source for the whole sweep. An attenuator in front of the receiver lowers the pass-band level, but it lowers the stop-band signal by the same amount, so it does not help the floor.

Lever 2: narrow the IF bandwidth

Noise power is proportional to bandwidth. The IF bandwidth is the width of the filter inside the receiver. Narrowing it by a factor of 10 lowers the noise floor by 10 dB:

ΔPfloor=10log⁡10B2B1\Delta P_{\text{floor}} = 10\log_{10}\frac{B_2}{B_1}

With B_2 = B_1 / 10 the change is −10 dB. The signal level does not change, so the dynamic range improves by the same 10 dB.

The price is time. A narrower IF bandwidth means a slower sweep. For a single deep stop-band check it is usually worth it. For a quick pass/fail sweep it may not be.

Averaging also reduces the scatter of the noise on the trace. Use it to make the trace smoother, but plan on bandwidth and source power for the real gain in floor.

Worked example

The numbers below are assumed for illustration. They do not describe any instrument.

Assume a source power of 0 dBm and a noise floor of −90 dBm at the starting IF bandwidth.

  1. Dynamic range: 0 − (−90) = 90 dB.
  2. Check the ratio: a level of −90 dB is 10^-9 of the source. 0 dBm is 1 mW, so the received power is 10^-12 W, which is −90 dBm. This equals the floor.
  3. Suppose the filter truly rejects 100 dB. The signal at the receiver would be −100 dBm, which is 10 dB below the floor. The trace shows the floor, near −90 dB, noisy and flat. It does not show 100 dB.
  4. Narrow the IF bandwidth by 10. The floor drops to −100 dBm and the dynamic range is 100 dB. The signal is now at the floor, so the trace is still not trustworthy.
  5. Raise the source power by 10 dB to +10 dBm, assuming the receiver does not compress. The stop-band signal is now −90 dBm, and the floor is −100 dBm. The dynamic range is 110 dB, and the stop-band reads 10 dB above the floor.

Each step of 10 dB comes from one lever. Both levers add.

A reading within a few dB of the floor is not a rejection value. Treat it as "at least this deep" and improve the dynamic range before you quote a number.

Other things that limit a deep reading

Dynamic range is the main limit, but not the only one. A deep stop-band trace can also be limited by leakage between the test ports and by cable and connector problems. If the trace sits above the floor and does not respond to bandwidth, suspect these. Good connectors and a calibration at the end of the cables help. Level conversions between dBm and watts are covered by the converter linked below.

In plain words · for students and beginners

Dynamic range in plain words

An everyday picture

Think of a quiet whisper in a noisy room. If the room noise is loud, you cannot hear the whisper. You can make it audible in two ways: the speaker talks louder, or you cut the room noise.

A VNA is the same. The "whisper" is the signal that gets through a filter's stop-band. The "room noise" is the noise floor of the receiver. The loudness of the speaker is the source power.

Where the picture breaks down: in a room you cannot choose how much noise you hear. A VNA lets you cut the noise by narrowing the IF bandwidth, which is the width of the window the receiver listens through. A narrower window lets in less noise.

0 dBm-10 dBmInput+10 dBmAfter amplifier (+20 dB)+7 dBmAfter cable (-3 dB)
Gains and losses in dB add to a level in dBm.

The same example in simple steps

Use the assumed numbers: source 0 dBm, noise floor −90 dBm. (dBm is a power scale where every +10 dB is ten times the power.)

  1. The gap between the two is 90 dB. That is the dynamic range.
  2. A filter that really blocks 100 dB sends a signal 10 dB below the floor. You cannot see it.
  3. The trace stops at the floor and looks like flat, noisy fuzz.
  4. Make the IF bandwidth ten times narrower. The floor drops 10 dB.
  5. Raise the source by 10 dB, if the receiver stays linear. The signal goes up 10 dB.
  6. The total gap is now 110 dB, and the filter's true rejection can be seen.

Mistakes beginners make

  • Reading the flat part as the answer. The fuzzy flat part is the floor. It is not the filter's rejection.
  • Raising the source power without limit. Too much power makes the receiver compress, and the pass-band readings become wrong.
  • Forgetting the cost of a narrow IF bandwidth. The sweep gets slower. Narrow it only as far as you need.
  • Trusting a trace that does not change. If the level does not move when you change the bandwidth, look for another cause, such as leakage or a bad connection.

Remember

  • Dynamic range is the source power minus the noise floor, in dB.
  • Narrowing the IF bandwidth by 10 lowers the floor by 10 dB, and raising the source by 10 dB gains 10 dB, until the receiver compresses.
  • A noisy, flat stop-band trace shows the instrument, not the filter.

Frequently asked questions

What limits the smallest S21 a VNA can measure?

The receiver noise floor and the source power. The transmission dynamic range is the source power level minus the noise floor, in dB. An S21 below that difference cannot be read.

Why does my filter's stop-band look flat and noisy?

The trace has reached the receiver noise floor. A flat, noisy curve there is not the true rejection of the device. The real rejection is at least as deep as the floor shows.

Does a narrower IF bandwidth improve dynamic range?

Yes. Noise power is proportional to bandwidth, so narrowing the IF bandwidth by a factor of 10 lowers the noise floor by 10 dB. The cost is a slower sweep.

Can I just raise the source power to see deeper rejection?

Raising the source power by 10 dB raises the measured signal by 10 dB, so the dynamic range improves by 10 dB. This holds only until the receiver starts to compress.

References and further reading