concerning digital interface sound quality include differences between optical links and wired coaxial connections, and changes in sensitivity to interface quality depending on DAC architecture. Is the digital audio interface flawed? Specifically, how can these claimed subjective differences occur in a digital data link? After all, "bits is bits." THE DIGITAL AUDIO INTERFACE STANDARD The AES/EBU and digital interface standards use biphase-mark encoding to transmit two-channel audio data, synchronization information, and subcode data over a single serial information channel;3 this coding scheme allows clock information to be embedded in the serial datastream. Fig2 shows the serial subframe structure consisting of 32-bit cells, each subframe carrying code for one audio channel.
The subframe begins with a 4-bit synchronization signal "preamble" followed by a 4-bit auxiliary data block. Up to 20 bits of audio data can be transmitted, with LSB (least significant bit) first, and the MSB (most significant bit) occupying the last audio cell position. Finally, subcode information comprises validity, user, channel status, and parity bits. The biphase-mark encoding technique places cell transitions at the beginning and end of each cell for "0" bits, and at the cell's beginning, midpoint, and end for "1" bits. The preamble violates this coding rule, so that interface receiver circuitry can detect when each subframe begins. If the audio data sampling rate fs = 44.1kHz, then the cell (0) width is equal to 354 nanoseconds, while the half-cell width (1) is 177ns; hence, the maximum rate of transitions is equal to 1,000,000,000/177 - 5.65MHz, though harmonics of the interface signal will extend to far higher frequencies. Fig.3 shows time-domain simulation of a single subframe carrying a left-channel audio sample of value 255, equal to 1111111100000000 in 16-bit, twos-complement notation with MSB last. The mid cell transitions can be seen at each "1" bit position, while biphase-mark violation displaces local transition positions in the preamble. 3 AES3-1985, "AES Recommended Practice for Digital Audio Engineering - Serial Transmission Format for Linearly Represented Digital Audio Data," JAES, December 1985, Vo133, pp.979-984. |
The biphase-mark signal can be transmitted using either a coaxial or optical connection, while the interface decoder at the receiver has to extract clock and audio data, and sub-code information, from the serial datastream. The clock signal embedded in the serial datastream is usually used to control a phase locked loop (PLL), which in turn should provide a stable reference frequency for conversion circuitry interfaced to the analog world. A number of dedicated Audio Digital Input Circuit (ADIC) integrated circuits now available will perform these functions. The circuit in fig.4 uses the Philips SAA7274 ADIC; negative going edges on the S/PDIF input signal are detected and compared to edges on the system clock derived from the PLUs 11.2896MHz crystal oscillator. A difference signal is fed to a varicap diode, which pulls the PLL oscillator frequency to match the clock frequency embedded in the incoming interface signal. The PLL has a first-order loop filter with a break frequency of approximately 1kHz, allowing clock recovery to reject short term variations in the input frequency (ie, high frequency jitter). When the interface decoder supplies data to a DAC, the analog audio output will be corrupted if the samples are the wrong value (amplitude or "bit" errors), or are output at the wrong times (jitter). AMPLITUDE ERRORS IN THE DIGITAL AUDIO INTERFACE The unfiltered digital interface waveform is a binary signal whose transmitted information is determined by the transitions in the signal. One of the benefits of biphase-mark encoding is that the interface signal has only a small DC component allowing interface signals to be AC coupled, and edge detection to be performed using a comparator referenced to ground. If an audio data-cell transition is missed at the receiver, a bit error occurs, and a DAC connected to the receiver will output an incorrect sample value. We now proceed to model a bandwidth limited link by filtering the subframe signal with a first-order (RC) low-pass filter, and determine what degree of filtering will result in bit errors. Using the first-order filter model is a gross simplification of the time-domain behavior of a real link - accurate analysis requires the use of transmission-line theory at the high frequencies involved - but it's a good starting point for investigation. Consider the top section of fig.5, which shows a simulation of the subframe signal carrying an audio word value of 255 and filtered using a time constant of 100ns, correspond- |
to previous page Stereophile, March 1996 to next page |