Feature Based Control of Compact Disc Players

Several new types of optical disc standards have reached the market during the past years, from the well known Compact Disc (CD), and its computer version CD-ROM, through the Digital Versatile Disc (DVD) to the new high density discs, eg. Blu-Ray discs. All these standards have potential problems with surface faults like scratches and fingerprints. Many users have suddenly encountered the problem that a CD-player, a PC or a DVD-player could not play a given disc, since this disc has become too faulty. The main topic of this thesis is to improve the CD-players’ capability to play these discs with surface faults, by improving the two controllers which keep the CD-player positioned on the CD. Some parts of this Ph.D project has been specialised for a specific CDplayer. However, the achieved results can be generalised to the other optical disc player standards. The problem in handling these surface faults is that they result in a faulty signal component in the position signals which are used for positioning the CD-player. The basic idea behind this study is to use feature extraction to remove this faulty signal component from the measured position signals. This Ph.D study contributes with six major technical contributions, which are all described in this thesis. Traditionally the optical parts of CD-players are modelled with simple linear models. However, these models are not ideal for detecting and handling surface faults. Instead a more detailed non-linear model of the optical part of the Compact Disc player is developed. Based on this new optical model and a simple model of the surface faults, a pair of residuals are defined. These residuals have shown better fault detection properties compared with the normally used residuals. In order to compute these new residuals a method to compute the inverse map of the non-linear optical model is introduced. Using this inverse map together with a Kalman estimator, the residuals can be estimated as well as an estimate of the position signals which are more valid during surface faults. Based on the residuals, time-frequency based methods are designed for detecting the surface faults. The results, however, have shown that a better strategy is to improve standard thresholding algorithms. Time-frequency based methods are used to extract features from the surface faults. First of all the faults are classified among different fault classes, and even more importantly, the faults are approximated by the use of timefrequency based methods. These approximations are used for two purposes. The first