100 Things #68: Using Transfer Function to Measure Smart Devices
Modern devices are becoming better and better at filtering out unwanted background noise from calls. This includes noises like sine sweeps, typically used to test these devices. Instead testing these devices with real world signals like speech, music, and noise can bypass noise suppression. The ability to use transfer function in SoundCheck to test these devices with non-linear signals was added in 2005. Transfer function can do more than just measure smart speakers, with the ability to also test loudspeaker impedance or compare measurements of a reference mic to a DUT mic, all with the same SoundCheck module.
Using Transfer Function to Measure Smart Devices
Learn more about transfer function, and other SoundCheck features
Read on about SoundCheck features and functionality, discussing algorithms, automation features, and more. If you’re ready to start testing on your own SoundCheck system, see our full catalog of free Loudspeaker test sequences.
Video Script:
Modern devices such as mobile phones, smart speakers, TWS earbuds and audio infotainment systems use sophisticated DSP algorithms to improve the voice quality and intelligibility of these devices. While these algorithms greatly improve the user experience, they create challenges for the audio test engineer as they often filter out common test signals such as sinewaves.
So, if we can’t use sinewaves, then what are our alternatives? SoundCheck’s transfer function algorithm, which was first introduced in Soundcheck back in 2005, can use broadband signals such as speech, music and noise as a stimulus to measure the frequency response of a device.
The transfer function algorithm works in the frequency domain to perform a complex FFT (including magnitude and phase) on the stimulus and response waveforms. From these two spectra, a variety of results can be calculated including frequency response, non-coherent distortion, coherence, non-coherence, signal to noise ratio, cross spectrum, coherent power and non-coherent power. Time domain analysis outputs are also available and include impulse response, auto-correlation of the stimulus and response as well as cross-correlation. A power averaging option can be used when the response waveform contains jitter or other artifacts that would result in phase calculation errors when complex averaging is used. When power averaging is selected, it limits the algorithm’s output to frequency response, auto spectrum of the stimulus and response waveforms and auto-correlation.
Let’s say we want to measure a microphone embedded in a smart speaker which we know uses DSP to filter out sinewaves. We construct a compound stimulus containing the smart speaker’s wake word, Alexa for example, followed by a pink noise stimulus having a minimum bandwidth equal to that of our device under test. When this stimulus is played from SoundCheck, the wake word triggers the smart speaker and its recording of the stimulus playback can be retrieved from the cloud as a WAV file and recalled into SoundCheck for analysis.
Transfer function analysis isn’t limited to just open loop measurements using non-sinusiodal stimuli. We can use transfer function to make high accuracy electrical impedance measurements by measuring the speakers terminal voltage and current or even use the recorded time waveforms of a reference mic and DUT mic to analyze the response of the DUT mic.
In this video, we’ve only scratched the surface of the capabilities of this powerful algorithm. If you’d like to learn more about it, contact one of our sales engineers to arrange a demo. Thanks for watching!
