Audio test CD

For repairing amplifiers.

Left and right channels have identical signals. The files were created using the excellent shareware Goldwave audio editor.

Track listing is as follows:

1) 1000Hz sine wave
2) 1000Hz square wave
3) 1000Hz triangle wave
4) 440Hz sine wave
5) 20Hz sine wave
6) 20KHz sine wave

Each track is only 1 minute long, so use “Repeat” on your CD player to obtain a (nearly) continuous tone.

Don’t play this CD through your speakers, unless the volume is at minimum.

The audio tracks are recorded at 50% dynamic range, to avoid any clipping in the CD player itself (of course it should not do that, but my portable CD player used for testing is not perfect, maybe yours isn’t either).

Download  a ZIP of the .ISO and .CUE file ready for burning with your favourite CD writing software.

Download a ZIP of the original WAV files.

XMOS StartKit - write performance of Sandisk memory card

Here's a graph showing the write performance to a Sandisk Class 10 memory card.

The reason for the sharp dips in performance, are that the FAT filing system has to go off and allocate a new cluster every N writes.  For example 512 byte sectors, 32Kbyte clusters, means a new cluster is needed every 64 writes.

Another factor on the XMOS board is whether the code is running in 'debug' or 'release' mode. This can make around 2x difference.

I also found that the code published on https://github.com/xcore/sc_sdcard seems to not pay attention to setup and hold times, and appears to be leaving the clock line in a logic "1" state when in fact data gets clocked on a logic 0 -> 1 transition.

All this led me to branch the repo and produce my own version, here: https://github.com/tuck1s/sc_sdcard

The changes are too numerous to list here (see the github changelog) but with pre-allocation of a file with contiguous sectors, and a Sandisk Extreme Pro SDcard, there's enough performance to write around 3Mbytes/s to the card.

My experiments with the even smaller "PetitFAT" filing system were halted because it only reads/writes a sector at a time and that would not give enough performance.

Measuring 'where the time goes', I can see that the time taken to clock in a sector's worth is usually only a few hundred microseconds, whereas the card can then take several milliseconds to respond that it's done. This is highly variable and also highly dependent on the grade of the SD card.