The Intel AudioScope Measurement System

Author(s):
Stuart Sherlock - Intel Corp.

Industry:
Telecommunications

Products:
LabWindows/CVI

The Challenge:
Developing a consistent resolution and tuning process to ensure audio clarity and overall quality of Intel telephony products.

The Solution:
Using National Instruments dynamic signal analyzer (DSA) boards to create a test process for audio performance designed to measure audio latency, data loss, volume, and frequency response.

Introduction
In 1996, when Intel Corporation developed the Intel Video Phone, audio performance of the media conferencing application was measured with subjective listening tests. As a result, the resolution and tuning process was inconsistent, and trial and error adjustments were often made. To ensure audio clarity and overall quality, Intel engineers developed an objective test-controlled process of audio performance. Audio performance encompasses latency and data loss due to transmission delays over Internet media or a public switch telephone network (PSTN) in media conferencing applications. Thus, the Intel audio measurement system was designed to measure audio latency, data loss, volume, and frequency response.

The Intel AudioScope Measurement System
AudioScope, the measurement system developed by Intel engineers, monitors an input audio signal from a sending system and the received output audio signal from a receiving system. To reproduce the audio inputs and to control the audio acquisition process, the system uses three standard types of wave files for repeatable reference sources; continuous, alternating, and pink noise. At the beginning of each type of wave file, a 0 dB amplitude, 10 cycle, 220 Hz sine wave signal is added to trigger the acquisition forlatency and data loss measurements. This burst signal is used to measure the audio delay times between the two client systems and to synchronize the received audio signal with the reference audio signal for data loss analysis.

Delivering unparalleled versatility and performance for audio measurements ranging from 20 Hz to 20 KHz, the acquisition system features a Pentium ® processor-based computer with National Instruments NI 4551 PCI dynamic signal analyzer (DSA) as the core measurement device. The NI 4551 onboard FFT signal processor delivers fast spectrum analysis and transient event analysis of real-time sampled time domain waveform inputs. Additional features of the system include data logging, presets, and a strip recorder.
We developed an intuitive virtual panel that interfaces with the NI 4551 board with Microsoft Visual Basic and National Instruments Measurement Studio tools for Visual Basic, a suite of ActiveX controls for measurement and automation applications. The user interface provides the controls and visual indicators for the test software. Menu items provide access to most of the program controls and functionality. With AudioScope options setting, the user can easily acquire other types of waveforms for real-time display and analysis such as volume verification, CODEC analysis, and jitter analysis.

Measuring Audio Latency
Audio latency, the amount of audio transmission delay in a call between the sending system and the receiving system, is measured with the NI 4551 from the time audio enters the sending system microphone (or sound card microphone input) to the time audio is received at the receiving system speakers (or sound card speaker output). The input reference signals are modified to support triggering on AudioScope received input channel. Thus, the acquisition is a point-to-point black box timing measurement. The measurement process is independent of the media used to support the audio call.

Measuring Data Loss
Data loss is perceived as a temporary loss of volume, usually associated with dropped audio packets. However, as network jitter increases, the audio playback buffer on the receiving system tends to under run. Audio artifacts begin to appear in a manner that simulates data loss. In actuality, there is no true data loss, just a change in the timing of the playback buffers. Rather, the timing of the playback buffers has been altered. A plot of playback audio suffering from jitter looks very similar to one with data loss. However, there are distinct pieces of the waveform missing when compared with the original input signal. For test purposes, both conditions can be plotted on the same graph to observe the results.

We designed and set AudioScope for the maximum displayed acquisition time base window of five seconds. This window is sufficient for latency measurements with synchronous triggering tones integrated in the audio wave files. However, analyzing the data loss acquired in this five-second window is difficult because the loss would occur randomly at any occurrence and duration. For this reason, AudioScope strip recording mode provides continuous acquisition over a user prescribed time, only limited by the size of the hard drive. The strip recorder records the transmitting reference audio signal and the received audio signal waveforms. The user can then zoom and pan the plotted data to analyze data loss events. The user visually inspects the plot using Audioscope tools to identify the magnitude and frequency of data loss. Markers are available to improve the latency analysis and to align the two waveforms due to latency for data loss analysis.

Peak-based envelopes are created and plotted for each transmitted and received signal. These two envelopes are then compared for differences. A summary envelope is calculated from the two envelopes and plotted over the result plot. The result and summary plots can support the user in determining the size and duration of any possible data loss events. The resolution of the summary envelope is set to correspond with 25 percent of the duration of the audio buffer size. The audio buffer size is determined by the particular protocols and CODECs used by the device under test.

Measuring Frequency Response
We perform real-time analysis using the NI 4551 fast fourier transform on the acquired time domain signals to perform frequency response measurements on CODECs. Frequency response is the variation of the system’s output level relative to the input level across a frequency range. With the typical human hearing range between 20 Hz and 20 KHz, audio systems with a flat response over this range seem to sound the most natural. As a performance benchmark, an in-band frequency flatness tolerance is usually specified in dB. Typically, the conventional tolerance was ±3.0 dB, which corresponds to the frequencies at which the signal voltage was reduced to one half the original signal level.

Calibration and Accuracy
To preserve the integrity of the acquired data, the system is calibrated by comparing signals with known latency values. We created a calibration wave file that is identical in both the left and right audio channels. We added the same 0 dB impulse signal to the beginning of each channel to give the test a distinct signal to fire the trigger and use it as the measurement point for both channels. The wave file to the second channel is then shifted in time by fixed increments. The wave file is played in a looping mode, and the latency is measured. The latency measured should match the known offset of the two wave file channels. Calibration is checked at several different latency intervals (100 ms, 200 ms, 400 ms, and 800 ms). We correct deviations from expected values by fine-tuning a scaling constant in the software.

Mobile Test System
We constructed a second system to meet the audio testing demands from other test and development groups within Intel. We organized this system to fit on a four-wheeled rack. The test systems are identical, with the exception of the sizes of the isolation boxes.

For more information, contact:

Stuart Sherlock

Senior Test Engineer

Intel Corporation

Tel: (503) 264-3089

E-mail: stuart.sherlock@intel.com

Browse All Case Studies »