How to Use Confidence Limits in SoundCheck

Balance speed and accuracy on your electroacoustic production line using SoundCheck’s confidence limits tools. These tools provide a clear visual display of measurement confidence, which helps you fine-tune your stimulus volume, cycle count, step duration, and frequency range to ensure accurate results with fast throughput.

Additional Resources for SoundCheck

Detailed explanations of how to use SoundCheck can be found in the SoundCheck manual.

Video Transcript: Measuring Confidence Limits

On an electroacoustic production line, faster testing means higher throughput. This makes it desirable to use a short test sweep. However, longer averaging time improves noise rejection leading to more accurate measurements, so how do you reach a good balance between speed and accuracy to get reliable results?

If you are using a stepped sine sweep in SoundCheck, our harmonic track analysis algorithm includes tools to help you find the best combination of speed versus accuracy, like confidence limits, standard error, and noise. Confidence estimates the signal to noise ratio inside the stepped sine filter; the longer the averaging time, the narrower the filter and hence the greater the measurement signal-to-noise ratio.  

I’m going to imagine I am testing a loudspeaker on a production line, and I want to get accurate results with the shortest sweep possible.

Let’s check all of these boxes and make a measurement so that we can see how it works.

First let’s take a look at our stepped sine stimulus. Several settings affect its length. I’m sweeping from 20 kHz to 20 Hz using a step sine sweep with R40 resolution, or 12th octave. At each step, I’m sweeping for a minimum of three cycles. The background noise, the signal-to-noise ratio, and the speed of the sweep all affect the measurement confidence. Let’s run this sequence and take a look. We’ll play the stimulus, acquire the data, and analyze it to get a frequency response.

Using tools from harmonic track analysis, I can now look at the fundamental upper and lower confidence limits. These measurements are saved in the memory list, so if I put these on the same graph as my frequency response, I can see that both the upper and lower confidence limits diverge from the frequency response at high and low frequencies. 

We can also look at the total noise of the measurement. In this graph, if I use a different color, I can see that at the low end, the environmental noise is actually greater than the measurement so our signal to noise ratio is leading to inaccurate results. 

So how can we improve this?

We can try to improve the signal to noise by playing the sweep at a louder volume. I can make sure that I have enough input gain to have a good signal to noise ratio using the max FSD in to make sure we’re at least minus 30 dB. I can also increase my stimulus cycles to reduce the influence of the noise. 

Now let’s look at my stimulus step. I’m sweeping from 20 kHz to 20 Hz at R40 resolution, with each step lasting at least three cycles. SoundCheck allows you to define these independently. Here I have 121 steps from 20 Hz to 20 kHz, and each step lasts at least three cycles. I can also set a minimum duration for each step, and the sweep will satisfy both criteria. Increasing the number of cycles or the minimum duration lengthens each step, but improves noise rejection.

If I increase the number of cycles to 20 and set a minimum duration of 20 milliseconds, it will transition at 20 cycles or 20 milliseconds, whichever is longer. At 1000 Hertz, 20 Cycles equals 20 milliseconds so for anything lower 20 Cycles is longer than 20 milliseconds and for anything higher 20 milliseconds is greater than 20 cycles. For example at 20 KHz, 20 Cycles is a very short amount of time so the minimum duration will increase the amount of Cycles to last 20 milliseconds.

Now you can see our original sweep is 2.7 seconds but let’s update it, and now you can see it takes 18 seconds, which might be too long for a production line. To keep it shorter we can make other changes like change the step resolution of the steps from R40 to R10, so the frequency range is the same but we have less steps per octave. Now if we update we can see it shortens the sweep to 5 seconds.

We can also adjust the frequency range based on what we really need to test our device. If we don’t really need to test down to 20 Hz or up to 20k, we could reduce the range to say, 100 Hz to 10kHz.  Now you can see this brings the sweep length down to 1 second, which is much more appropriate for a production line test. 

Let’s run the sequence again and see what our measurement confidence looks like with these new settings. We can see now the confidence is almost exactly where the frequency response is. It’s diverging a little bit in the low frequencies due to high total noise and because the speaker’s rolling-off a little near 100 Hz. 

Let’s try one more tweak. Let’s try 10 cycles instead of 20. Now the sweep is only 680ms long. And if we run it, we can see that the confidence is much less, so maybe we do need to increase the cycles and go with a slightly longer stimulus.

We can continue to optimize our stimulus step using these tools to achieve the best combination for our speed and accuracy requirements. 

And one last thing… if your measurement confidence is low due to noise, don’t forget to try to reduce your background noise – this will probably have the biggest effect on your measurement confidence!