Efficient realization of biorthoganal [sic] fixed point cosine modulated filter banks

Cosine modulated filter banks are widely researched and used in signal compression because they can provide perfect reconstruction (PR) while maintaining a low computational cost. Their ability to provide PR can be hindered by the device that they are implemented on. Fixed-point digital signal processors require that the filter coefficients be quantized, force resultants to be rounded, and are subject to overflow. All three can lead to reconstruction errors. For the filter bank to preserve its PR properties, its realization structure must be robust to these three sources of error. This structure is then implemented on a fixed-point device, such as a TI C6201 DSP without losing a valuable property. This is validated in this thesis along with showing that the frequency response of the filters is upheld by this structure through the use of Matlab simulations supported by a DSP implementation