Sub-Band Processing of Synthetic Aperture Sonar Data Sérgio Rui Silva 1 , Sérgio Cunha 1,2 , Aníbal Matos 1 , Nuno Cruz 1 1 Porto University, Faculty of Engineering (www.fe.up.pt), 2 CIMAR (www.cimar.org) Email: {srui, sergio, anibal, nacruz}@fe.up.pt Abstract- High frequency synthetic aperture sonar systems require demanding tolerances in motion errors and medium phase stability. This article proposes a new method that mitigates the problems associated with small wavelength related errors. By dividing the received signal bandwidths in to several smaller ones and conjugate complex multiplying them, a new signal is obtained with longer effective wavelength, thus reducing the impact of motion errors and medium phase fluctuations. Keywords- Synthetic aperture, sonar, signal processing. I. INTRODUCTION Today’s sonar systems strive for higher and higher resolution. This utterly leads to the transmission of higher frequency signals. Besides the obvious impairments on using high frequency signals such as signal attenuation, unknown platform motion errors and stability of phase propagation through the medium become increasingly important. As an example, the synthetic sonar system developed at University of Porto ([7]) uses a transmission frequency of 200kHz (wavelength of 7.5mm) and a useful bandwidth of 30kHz. This requires having its unknown motion errors below at least 1mm to enable accurate synthetic aperture image formation ([1, 2]), after auto-focus. It requires motion errors bounds prior to auto- focus to be in the order of magnitude of the wavelength, which is by itself quite demanding. This article proposes a new technique that mitigates the problems associated with the small wavelength of high frequency sonar signals. By dividing the received signal bandwidth in several smaller bands and conjugate complex multiplying the pulse compressed signals obtained in each band one by the other, a new resulting signal is obtained with an effective longer wavelength corresponding to the frequency difference between the two sub-bands. This longer wavelength effectively reduces the impact of phase fluctuation from the medium and platform motion uncertainties. Processing the signal trough this technique can be used to obtain directly a coarse final synthetic aperture sonar image, or used in an global contrast optimization auto-focus algorithm ([3, 4, 7]) with several steps with increasingly smaller wavelengths. This enables faster convergence through the resulting smother cost function surfaces (less impact of local minima). If a direct image in intended, a resulting loss of range resolution is unavoidable (due to the use of smaller effective bandwidths) and the along-track resolution even further affected (due to the use of a significantly larger effective wavelength while maintaining the path length covered by the sensor while scanning each target point). Figure 1: Autonomous boat in operation in operation in the Douro river, Portugal. This is enough, however, to auto-focus the synthetic aperture image data without the need for any special image features for a system with only one transducer and compensate for rough navigation data, prior to process the image at the native wavelength. In applications where rougher images are acceptable, this technique can be preferable to using lower frequency transducers directly. Although lower frequencies transducers can produce better along-track resolutions if larger apertures are employed (and this can be difficult in the terrain), their cross-track resolution is easily worse: the bandwidth of lower frequency transducers is significantly worse than that of higher frequencies, even after dividing the bandwidth into sub-bands. A common RTK differential GPS system has an error in the centimeter level. By using two 15kHz sub-bands within the 30kHz bandwidth of the transmitted signals (with a center frequency of 200kHz), an equivalent wave of 15kHz. This corresponds to a wavelength of 10cm, which is larger than the navigation error. For this wavelength and navigation accuracy, correct image formation with auto-focus algorithms are very efficient (only minor phase adjustments are needed for motion compensation). Using an auto-focus algorithm directly with the