KEYSIGHT U8903B AUDIO ANALYZER
SWEPT HARMONIC-DISTORTION INVESTIGATION

BENCHNOTES BN2020-2

Area Focus

Buildera BENCHNOTES™ BN2020-2 documents the Keysight U8903B audio analyzer harmonic distortion at low frequencies between 10 Hz and 110 Hz, and the influence of FFT sample size on distortion-measurement accuracy.

Background of Investigation

As part of a vibration-analysis investigation using piezoelectric and MEMS sensors, Buildera conducted a series of low-frequency harmonic-distortion measurements below 110 Hz and observed unexpected results at very low frequencies. Namely, the swept harmonic distortion level between 10 Hz to 30 Hz was far higher than published specifications, initially stirring speculation that the generator or analyzer required repair. Buildera took multiple measurements to compare and understand the measurement behavior, which was later presumed to be artifacts of the default FFT resolution setting (32768 samples). Failure to understand this relationship could result in materially erroneous measurements with errors in excess of 100 dBV, particularly at frequencies below 100 Hz.

Test Setup

The test set up included the following equipment:

Device Under Test (“DUT”)

• KEYSIGHT® U8903B Audio Analyzer

Supporting Test and Measurement Equipment

• None required

Instrument Set Up

To establish an initial reference measurements, BUILDERA set the U8903B as follows:

Analog Analyzer Settings

• PARAMETER: Frequency

• START FREQUENCY: 10 Hz

• STOP FREQUENCY: 110 Hz

• POINTS: 101

• SPACING: 1.000 Hz

• DWELL: 10.00 ms

• SRC CHANNELS: 1

• MEAS CHANNELS: 1

AG1

• AMPLITUDE: 1.0 V(rms)

• FREQUENCY: 1 kHz

• WAVEFORM: Sine

• DC Offset: 0.0000 V

• CONN: Loopback

• IMP: 50 Ω

• GROUND: Float

AA1

• TRACE 1: THDL (dBV)

• TRACE 2: AC Voltage 0.0000 V

• LPF, HPF, W, DeEM: None

• CONN: Loopback

• DC Offset: 0.0000 V

• CONN: Loopback

• IMP: 200 kΩ

• COUP: AC

• BW: 96 kHz

Measurement Results

DUT MEASUREMENTS

To avoid contaminating measurements with external DUT uncertainty, the loopback capability provides a direct internal connection between the output signal (AG1) and the audio analyzer input channel (AA1).

What Do The Measurements Show?

1. With the Kaiser window enabled, the measurements show that at least 262164 samples are required to reliably measure THDL at frequencies as low as 10 Hz. However, one tradeoff is that increasing the number of samples exponentially extends the measurement time in tandem. This may be acceptable for some R&D situations where measurement speed is less critical, but not for production where fast measurements are critical to throughput. The maximum limit is 2M (2097152) samples, or 2^21. Note that the instrument is capable of generating and analyzing frequencies as low as 5 Hz. Buildera selected a measurement range between 10 Hz to 110 Hz such that each grid line equals an even 10 Hz increment.

2. At the default setting of 32768 samples, the lowest reliable frequency for total harmonic-level measurement is approximately 70 Hz, below which the THDL fluctuates excessively, requiring a larger FFT sampling window.

3. The oscillation of THDL is unexplained in the transition region, but is likely due to anomalies from spectral leakage and the particular frequencies chosen. It appears that the FFT analyzer is not coherent with the generated frequencies, causing partial time-domain clipping of an integral number of waveform cycles inside the sampling window.

References

Keysight Technologies. Keysight | U8903B Analog Audio Analyzer - Demo Guide (5992-0667EN). 01 DEC 2017. Online: https://www.keysight.com/us/en/assets/7018-04782/configuration-guides/5992-0667.pdf

Chioye, Luis. Choose the Right FFT Window Function When Evaluating Precision ADCs. Electronic Design. 12-NOV-2013. Online: https://www.electronicdesign.com/technologies/analog/article/21798689/choose-the-right-fft-window-function-when-evaluating-precision-adcs

About BENCHNOTES

BENCHNOTES provide useful insights and tips discovered in the daily engineering and investigation process at BUILDERA. By sharing these findings online or in downloadable PDF format, engineers and consumers benefit from the exploration. BENCHNOTES are informal reports, often with raw, unaudited data.