Created on 2008-07-11 12:55:55
By Bruce Hofer (from a talk given to the AES-LA section, May 30th, 2007)
Increasing speed is a common goal in most technologies, and audio test is no exception.
It wasn't until the mid-1970s that audio test first started to experience significant improvements in speed: "auto set level" reduced distortion measurement times from 50 seconds per data point (if you were good) to less than 15 seconds. Then in 1980 Tektronix's AA501 became the world's first automatic audio analyzer with auto-ranging of all measurement functions, auto set-level, and digital numeric readout. The addition of GPIB ("HPIB") soon followed, enabling automated measurement sequences. With new advances in the HP8903 in 1980 and the Tektronix AA5001 in 1982, a 16-point 20-20kHz graph of THD+N that used to take 15 minutes could be created in less than one minute.
While automation decreases setup time and reduces the chance of human error, it does not impact actual measurement time. At the same time, point-by-point sine or stepped-sine analysis often misses unexpected problems and behaviors in very narrow bands that require greater resolution to be identified properly.
The need for faster and more thorough testing gave rise to several non-sinusoidal techniques over the years such as maximum length sequence (MLS) and multi-tone testing.
MLS is extremely fast, using a pseudo-random noise signal where every possible bit pattern occurs once during its repetition interval. A Hadamard transform of the output signal yields the impulse response of the device under test. MLS techniques are particularly attractive for acoustic measurements but their usefulness is compromised by non-linearity in the device under test, as non-linearity causes response errors that cannot be readily identified.
Multi-tone analysis offers attractive speed benefits and is useful when testing playback only devices such as MP3 players (the single tone can be stored on the device and played back just once versus having to repeat the acquisition for each step in a traditional point-by-point sine measurement). The test signal is a superposition of between 5-31 sine-waves from which measurements are calculated in a windowless FFT. The disadvantage of multi-tone is that it suffers an inherent lack of correlation with traditional techniques. The measurements aren't wrong, they're just different. This can be very confusing when trying to make comparisons using the different test techniques.
The newest approach to increasing the speed of test is swept-sine or chirp testing. While the concept of linear chirp (where frequency varies linearly with time) has been around for over 30 years, it has limited application especially in calculating THD+N measurements.
The value of chirp changed dramatically when Dr. Angelo Farina of the University of Parma introduced the first logarithmic chirp technique at the 108th AES Convention in 2000.
Dramatic improvements in speed were made possible because frequency increases exponentially over time: with a log chirp, the time to sweep from 20 Hz to 40 Hz equals the time to sweep from 10 kHz to 20 kHz
Farina showed that a deconvolution of the input and output signals yields time separated impulses of the linear and harmonic responses of the system being tested. By carefully time-windowing and transforming the various impulses, the individual response functions can be recovered from which many different measurements can then be derived mathematically. In other words, almost every common audio measurement can be obtained from a single acquisition.
Farina's technique was recently extended by Audio Precision to simultaneously test multiple channels through use of very small time offsets between channels (as seen in the new APx585 multichannel audio analyzer). This enables one to measure crosstalk in addition to response and distortion in multiple channels simultaneously. Incredible amounts of data can be extracted from a single, multiple-channel test.
New techniques will come along, but for now, chirp offers the most promising avenue for test speed. With the increase in computing power in today's PCs, we should see significant performance improvement year on year. It's an exciting time for audio test.