CD Transports: Data Drops Etc...

Followup to Aragain:

The CD system depends on the Nyquist Sampling Theorem, which states that a band-limited analog signal (22.05 khz for CD) sampled at at least twice the band limt (44.1 khz for CD) can be exactly reconstructed from the time series of samples.

A tacit assumption behind the Nyquist theorum is that the time interval between the samples be precise and equal to 1/fs, where fs is the sampling frequency (44.1 khz for CD). Any variation in the interval between samples AT THE DAC causes distortion (usually additional non integral high harmonics) in the reconstructed signal.

Thus recovering the data from the disc is only one part of the problem, and the easy part. That's why cheap CD rom drives have no difficulty extracting data from discs in computers. There is no time critical element involved in reading a file.

Audiophile CD transports have no problems getting the data either. It's the timing that causes problems. The clock signal usually starts at the transport (there are exceptions - see the Wadia 270 - 27ix combo). Any tracking problems from a warped disc may affect the stability of the clock at this point.

There are many opportunities for timing errors (jitter) to appear as the signal moves from transport to DAC. Almost all transport/dac seperates communicate using the SPDIF (Sony-Phillips Digital Interface) standard across coax or optical links. SPDIF combines the data stream and the clock together into one signal for transmission on one cable. The receiver at the DAC must extract the clock and data from the SPDIF. This extracted clock often directly becomes the timing reference to the DAC (though some processors go to elaborate lengths to stabilize this clock signal).

The SPDIF signal is itself an analog signal, and is subject to degradations in transit. Thses degradations can alter the waveshape so that the receiver has a hard time decting exactly when a clock transition occurs. This uncertainty is one cause of jitter, and is one reason that digital interconnects sound different.

I'm just scratching the surface here. Going back to T_bone's original note, one could say that the engineering is faulty. For example there are other bus schemes (I2S for example) that transmit clock and data on separate physical lines.

All things being equal (which they never are!) a single box player has an easier time of things since all components are under the designer's control and he/she doesn't have to use SPDIF to transfer data to the dac section.

Hi KThomas -

I'm in the computer game also and this same cd drive question has bugged me for a while.

One analogy is the old 'framing error' on a modem's uart where the uart's clock got out of sync with the sender's clock.

Still, it seems like this is a problem that has been solved already and could be built into a cd player.

I am trying to remember where I read a bunch of stuff about jitter. I suppose a web search would eventually turn up something insightful.

Cheers,
A

Ghost - Thanks for the info!

Still, this seems like a standard async communication problem (particularly in the single-box players) which has been solved in the computer data communications world. I think the serial rate on SPDIF is 2 mhz or 4 mhz.

If the dac is viewed as an async receiver (which I don't know if that's a valid view) then it seems existing engineering is available to get it right.

Still puzzled -
A

It's the framing error on a UART model that I have in mind for this data transfer too. Maybe that's the wrong mental model, and when I get the right mental model it'll be clearer to me. Certainly, one problem with digital audio playback is that these connections don't have redundancy built in (ie, no retry logic). However, I can connect two computers up via async ports (ie, UARTS) and write a simple program to send on one side, receive on the other, and report any chip-level errors, let it run for days and not see a single error, all without retry logic, etc. So, it still seems very solvable, at least to my level of understanding of the problem set.

I guess another way I think of it is this - what if somebody re-engineered the concept of a CD transport / DAC combination such that the transport didn't attempt to play it back real-time, but rather read the CD "in" and reliably loaded it into memory on the DAC. This would be akin to reading a computer CD into memory and, since you'd be re-engineering the interface, could base it on different technology. Then, the "CD" would play back from within the DAC's memory. I would think the whole issue of jitter would be moot with such a setup, and the whole notion of an expensive transport and expensive interconnects would be moot as well.-Kirk

Remember, that unlike an async serial data transfer between two uarts, the SPDIF signal contains both the data and clock streams intertwined. Unlike a uart, which knows a priori the operating baud rate, the SPDIF receiver must extract the clock from the incoming signal and then use this clock to derive the data stream from the signal.

Any distortions that round off the corners of the analog waveform making up the SPDIF signal cause the receiver to derive a jittery clock signal. In any digital logic, there is a transition zone between the voltages representing the logic states "1" and "0" that is neither state. A rounded off waveform spends more time in this transition zone, thus delaying the detection of logic transitions, thus affecting the derived clock signal.

This effect is not constant.

Improperly terminated coax (RCA plugs instead of BNC connectors, or bad transmitter/receiver design) will have reflections giving rise to standing waves which will cause distortion to the SPDIF signal.

Basic CD player design is still rooted 1983 technology. In those days memory and processor were expensive. I think that one could completely rethink the player design today, using current digital technology and its new price points.

Perhaps the CD drive (at 4x or better) could read ahead, and fill up a circular buffer. The data stream could then be clocked out of the buffer at a constant rate independently of the transport This scheme would even allow for retries of uncorrectable read errors, and would break the direct connection between data timing and transport timing. Of course we'd want to dump SPDIF as a connection mechanism...

I bet it would be easy to make using off-the shelf parts as well.