Toronto AES Bulletin April 1996 Meeting Review

---

Graham Boswell On Jitter In Digital Audio

One of the few experts on this subject, Graham Boswell of Prism Sound attracted an august audience of academics, CD production and digital audio professionals for what turned out to be one of the longest but most worthwhile meetings of the year. Graham's introduction to the topic was enhanced greatly by the full set of jitter analysis equipment he brought with him including: a DSA-1 Digital Audio Interface Analyzer, a JM-1 Jitter Modulator, a DA-1 D-A Converter, an AD-1 A-D Converter and an SNS Super Noise Shaper with a Dolch FFT Analyzer. He explained that the need to design and build jitter analysis gear arose in the late '80's when inappropriate cabling and slack practices resulted in audible digital signal degradation in the early CDs.

Graham described the work of Julian Dunn (also of Prism Sound) who analyzed the impact of jitter on digital audio. He suggested that in the digital audio path, a sinusoidal 500 psec p-p bit-timing error (jitter relative to its smoothed average) will cause sidebands, 96 dB down, around a 20 kHz signal. Masking will generally render these sidebands inaudible, but if the signal is in the 1 -10 kHz range, less than 300 psec of repeating jitter may be audible. At much lower frequencies the effect of jitter is much less audible.

A demonstration with a 12 kHz signal showed a 4 kHz spurious result, at 66 dB down (on the Dolch Analyzer), by introducing 22 nsec p-p sampling jitter. Halving the jitter to 11 nsec p-p dropped the spurious result by 6 dB, as expected. This helped to demonstrate that sampling and sampling-rate conversion processes require absolute constant rates, and that existing jitter must not be allowed to propagate.

There are many origins of digital audio jitter: unsteady clock oscillator due to crosstalk or dynamic loading effect; transmission cable due to poor impedance matching, crosstalk or noise pickup from unbalanced twisting; incorrect cable driving waveshape; incorrect cable-length compensation network; intersymbol interference (strings of ones and zeros having different transit times through a cable); and unstable operation of signal-recovery phase-lock-loop circuits, to name a few.

Special test signals are also available from Julian's test equipment to maximize intersymbol interference in a cable or a device under test allowing one to observe the "closing of the eye" at the end of the path on an oscilloscope, or the resulting spurious products on an analyzer. Graham feels this "stress test" should be performed routinely rather than as a last resort. He treated us to a demonstration by inserting a 100 meter piece of low-cost cable into a direct digital audio link. The 12 kHz signal came through fine, but the noise floor rose by about 22 dB. This is a very good new diagnostic tool for finding the weak spots in a system. As it turned out the problem here was exacerbated by a poor DAC. The Prism DAC with it's carefully designed PLL essentially eliminated the noise.

Prism Sound is proposing a required jitter tolerance for digital audio equipment since the acquisition of jitter in transmission is a given. The curve is at 10 digital bits of tolerance over the 1 to 200 Hz rate, decreasing to 1/3 bit at an 8 kHz jitter rate and constant at 1/3 bit. This reminds me of my days in 1.544 Mbit digital telephone where we worked out these very same problems about 16 years ago.

There is now an emerging consensus in the mastering community that CDs mastered and/or pressed at different facilities do not sound the same. There is however, little consensus as to why this is happening or even how it is possible. However Prism is setting up a program of systematic testing and measurement of the various properties of all popular CD recorders, up-scale players, and their controlling software/firmware in an effort to pin down the cause. Twelve test CDs are being made in several plants using different media and hardware. These CDs will be checked and verified for error, then used in double-blind subjective listening tests using a Prism Sound DAC as the reference. Ian Dennis and Julian Dunn will present their results in a paper to be given at the AES Convention in L.A. this November. An early finding is that low-limit speed of operation in a CD player (over 70 min of time) causes more problems with certain data sequences than at nominal speed.

Thanks to Peter Cook for organizing this event.

Review by John Fourdraine

| Preview of this Meeting | Back to the October 96 Issue |

| Toronto AES Home | Current Bulletin | Future Meetings | Back Issues | Search |