In this thesis, the JPEG 2000 quality scalability feature was investigated in thecontext of transporting video over IP networks. The goals of the investigation wastwo-fold. First, it was desired to nd a way of choosing the number of quality layersto embed in a JPEG 2000 codestream. In previous work, this choice has been moreor less arbitrary. Second, it was desired to nd how low the video bitrate could bedropped before it became perceptible to a viewer. This information can be usedin an IP networking scenario to e.g. adapt the video bitrate blindly according tothe measured channel capacity as long as the drop in bitrate is expected to beimperceptible. When the drop in bitrate is expected to be perceptible, a switchcould be made to a smoother bitrate adaptation.A way of choosing the total number of quality layers to embed in a codestreamwas found by minimizing the dierence in predicted quality between direct andscaled compression. Scaled compression is the compression which is achieved byextracting quality layers. The minimization procedure was bound by the speed ofthe encoder, as it takes longer for an encoder to embed more quality layers. It wasfound that the procedure was highly dependent on the desired bitrate range.A subjective test was run in order to measure how large a drop in video bitrate hadto be for it to become perceptible. A newly developed JPEG 2000 quality layerscaler was used to produce the dierent bitrates in the test. The number of qualitylayers to embed in codestream was found by using the minimization procedurementioned above. It was found that, for the bitrate range used in the test, 2 - 30Mbits/s for a resolution of 1280x720 at 25 frames per second, the magnitude ofthe drop in bitrate had to be at least 10 Mbits/s before the participants in the testnoticed it. A comparison with objective quality metrics, SSIM and PSNR, revealedthat it was very dicult to predict the visibility of the drops in bitrate by usingthese metrics. Designing the type of rate control mentioned in the rst paragraphwill therefore have to wait until a parameter with good predictive properties canbe found.

Pipes with corrugations or cavities are used in a wide variety of applications. In recent years the natural gas industry has experienced "singing" risers, where pipes transporting natural gas excite loud whistling sounds limiting the flow rate of which the gas can be transported. There has been a number of publications regarding this phenomenon, investigating corrugated pipes and pipes with one or more cavities. In this thesis the most basic situation will be studied; A smooth pipe with a single small cavity. This is studied by simulations and experiments. The effects of changing the length of the cavity, and the pipe section between the inlet and the cavity is investigated. The simulations were conducted in Palabos, a Lattice Boltzmann solver which proves to be a promising piece of software for acoustic simulations. The experiments were conduced using a metal pipe with variable inlet and cavity length. Initial vortices created at the inlet are amplified in the cavity by a cavity flow. The results strongly suggest that these inlet vortices are essential for the excitation of whistling sounds. The number of vortices traveling across a cavity at the time is called a hydrodynamic mode. When the frequency of vortices crossing a cavity coincides with an acoustic pipe mode, a whistling sound close to this frequency is excited. Cancellation of the whistling sound with an added cancellation frequency is possible for certain cavity and inlet lengths.

The NTNU Test Satellite (NUTS) is planned to have a payload for observation of atmospheric gravity waves. The gravity waves will be observed by means of an infrared camera imaging the perturbations in the OH airglow layer. So far, no suitable camera has been found that complies with the restrictions that follows when building a small satellite. Uncooled InGaAs has however been concluded to be the most suitable detector type in terms of wavelength response and weight.InGaAs sensors are known to have a high dark current when not cooled, and processing must therefore be applied to remove the background offset and noise.The combination of the high speed of the satellite and the long exposure time that is required for the camera will create motion blur. Simulations with synthetic test images in MATLAB showed that the integration time should at least be kept under 1 second in order not to destroy the wave patterns. Longer integration times may however be required in order to get a sufficient SNR.Two signal processing solutions to this problem was investigated: motion blur removal by deconvolution and image averaging with motion compensation. The former strategy is to apply a long exposure time to get a strong signal, and then remove the blur with deconvolution techniques using knowledge of the blur filter.Simulations applying the Lucy-Richardson (LR) algorithm showed that it was not able to remove strong blur, and was very sensitive to errors in the blur filter and noise in the image. The other approach is to obtain a sequence of images with short exposure time in order to avoid motion blur, and provide the necessary SNR by shifting the images according to the known motion and combine them into one image. This concept is simpler and more reliable than the deconvolution approach, and simulations showed that it is less sensitive to errors in the speed estimate than the deconvolution algorithm. It was concluded that this is the most suitable approach for the NUTS application, and it should be implemented on-board the satellite in order to provide a good SNR for the compression to function optimally. The downlink datarate of NUTS is of only 9600 bit/s, and it has been estimated that 2.45 Mb of payload data can be downloaded on average per day. This corresponds to less than 5 uncompressed images of 256 × 256 pixels with 8 bit per pixel.A sequence of overlapping combined images should be obtained to provide a scan of a desired area, and it was suggested that it should be encoded as video to enable efficient compression and transmission of as many images as possible to the ground station. A three-dimensional DPCM algorithm combined with a deadzone quantizer and stack-run coding was implemented in MATLAB. Simulations demonstrated that this simple compression scheme can provide a bit rate of less than 1 bit/px for a sequence of ravity wave images. One of the quantizers that was tried gave 0.83 bits per pixel with reasonable quality. If this number can be achieved in practice, the image transfer ate would be increased to 45 images per day, which is a significant improvement.

This thesis has taken a closer look at the implementation of the back-end of a language recognition system. The front-end of the system is a Universal Attribute Recognizer (UAR), which is used to detect phonetic characteristics in an utterance. When a speech signal is sent through the UAR, it is decoded into a sequence of attributes which is used to generate a vector of term-count. Vector Space Modeling (VSM) have been used for training the language classifiers in the back-end. The main principle of VSM is that term-count vectors from the same language will position themselves close to eachother when they are mapped into a vector space, and this property can be exploited for recognizing languages. The implemented back-end has trained vectors space classifiers for 12 different languages, and a NIST recognition task has been performed for evaluating the recognition rate of the system. The NIST task was a verification task and the system achived a equal error rate (EER) of $6.73 %$. Tools like Support Vector Machines (SVM) and Gaussian Mixture Models (GMM) have been used in the implementation of the back-end. Thus, are quite a few parameters which can be varied and tweaked, and different experiments were conducted to investigate how these parameters would affect EER of the language recognizer. As a part test the robustness of the system, the language recognizer were exposed to a so-called out-of-set language, which is a language that the system has not been trained to handle. The system showed a poor performance at rejecting these speech segments correctly.

CLIC is an international study of a future multi-TeV electron-positron linear collider, where the energy of a high-intensity drive beam is extracted and transferred to the main beam via Power Extraction and Transfer Structures (PETS) in the form of rf power. The study of power production is therefore essential for the feasibility of CLIC. Power production in PETS has been studied, and experiments have been performed in the decelerator Test Beam Line in the CLIC Test Facility 3.In particular, the correlation of the power production and the beam position inside the structure has been studied. It is shown that the total produced power is constant when the beam has a position offset through the PETS. In addition, the difference between the measured phases from each side is independent of the beam position, which allows for efficient combination of the fields. However, the ratio of the power on each side of the PETS unexpectedly shows a linear dependence on the horizontal offset, with a correlation value of 0.87. This can potentially affect the power transferred to the main linac, and should be taken into account in the design of the high power rf system.A graphical user interface was developed for the Test Beam Line, and the functionality is described in detail. The program is used in the operation of the Test Beam Line for monitoring, matching, steering and power production experiments, and has interfaces to MAD-X and PLACET. A documentation is also given of the Test Beam Line and the methods of the power measurements.

Denne Rapporten gir en forklaring på hva programvaredefinert radio er generelt. Den omtaler en konkret realisering ved hjelp av programvaren GNU Radio og den universale radioenheten USRP (Universal Software Radio Peripheral). GSM forklares slik at leseren får den nødvendige oversikten som trengs for å forstå resten av rapporten. Frekvenshopping forklares generelt og hvordan det er gjort i GSM. Det fremstilles en mulig realisering av frekvenshopping med USRP. Programvaren som benyttes for å opprette et GSM-nettverk presenteres. Programvaren som trengs er OpenBTS, GNU Radio og Asterisk. Den fysiske oppkoblingen av USRP når OpenBTS benyttes i programvare GSM blir forklart og hvilket testmiljø eksperimentene ble utført i beskrives. Delvis vellykkede resultater presenteres og rapporten konkluderer med at det skal være mulig å implementere frekvenshopping i OpenBTS med USRP som RF-maskinvare.