Analog-to-digital converters (ADCs) are required in almost all signal processing and communication systems. They are often the most critical components, since they tend to determine the overall system performance. Hence, it is important to determine their performance limitations and develop improved realizations. One of the most challenging tasks for realizing software defined radio (SDR) is to move ND conversion as close to the antenna as possible, this implies that the ADC has to sample the incoming signal with a very high sample rate (over 100 MSample/s) and with a very high resolution (14 -to -16 bits). To design and implement AID converters with such high performance, it is necessary to investigate new designing techniques. The focus in this work is on a particular type of potentially high-performance (high-resolution and highspeed) analog-to-digital conversion technique, utilizing filter banks, where two or more ADCs are used in the converter array in parallel together with asymmetric filter banks. The hybrid filter bank analog-todigital converter (HFB ADC) utilizes analog filters (analysis filters) to allocate a frequency band to each ADC in a converter array and digital synthesis filters to reconstruct the digitized signal. The HFB improves the speed and resolution of the conversion, in comparison to the standard time-interleaving technique by attenuating the effect of gain and phase mismatches between the ADCs. Many of the designs available in the literature are compromising between some metrics: design complexity, order of the filter bank (computation time) and the sharpness of the frequency response rolloff (the transition from the pass band to the stop band). In this dissertation, five different classes of near perfect magnitude reconstruction (NPMR) continuoustime hybrid filter banks (CT HFBs) are proposed. In each of the five cases, two filter banks are designed; analysis filter bank and synthesis filter bank. Since the systems are hybrid, continuous time IlR filter are used to implement the analysis filter bank and digital filters are used for the synthesis filter bank. To optimize the system, we used a new technique, known in the literature as frequency response masking (FRM), to design the synthesis filter bank. In this technique, the sharp roll-off characteristics can be achieved while keeping the complexity of the filter within practical range, this is done by splitting the filter into two filters in cascade; model filter with relaxed roll-off characteristics followed by a masking filter. One of the main factors controlling the overall complexity of the filter is the way of designing the model filter and that of designing the masking filter. The dissertation proposes three combinations: use of HR model filter and IlR masking filter, HR model filter/FIR masking filter and FIR model filter/FIR masking filter. To show the advantages of our designs, we considered the cases of designing the synthesis filter as one filter, either FIR or IlR. These two filters are used as base for comparison with our proposed designs (the use of masking response filter). The results showed the following: 1. Asymmetric hybrid filter banks alone are not sufficient for filters with sharp frequency response roll-off especially for HR/FIR class. 2. All classes that utilize FRM in their synthesis filter banks gave a good performance in general in comparison to conventional classes, such as the reduction of the order of filters 3. HR/HR FRM gave better performance than HR/FIR FRM. 4. Comparing HR/HR FRM using FIR masking filters and HR/IIR FRM using IIR masking filters, the latter gave better performance (the performance is generally measured in terms of reduced filter order). 5. All classes that use the FRM approach have a very low complexity, in terms of reduced filter order. Our target was to design a system with the following overall characteristics: pass band ripple of -0.01 dB, stop band minimum attenuation of - 40 dB and of response roll-off of 0.002. Our calculations showed that the order of the conventional IIR/FIR filter that achieves such characteristics is aboutN =2000. Using the FRM technique, the order N reduced to aboutN = 244, N = 179 for IIRJFIR and IIR/IIR classes, respectively. This shows that the technique is very effective in reducing the filter complexity. 6. The magnitude distortion and the aliasing noise are calculated for each design proposal and compared with the theoretical values. The comparisons show that all our proposals result in approximately perfect magnitude reconstruction (NPMR). In conclusion, our proposal of using frequency-response masking technique to design the synthesis filter bank can, to large extent, reduce the complexity of the system. The design of the system as a whole is simplified by designing the synthesis filter bank separately from the design of the analysis filter bank. In this case each bank is optimized separately. This implies that for SDR applications we are proposing the use of the continuous-time HFB ADC (CT HFB ADC) structure utilizing FRM for digital filters.

Thesis (M.Sc.Eng.)-University of KwaZulu-Natal, Durban, 2005.