Floating-Point Analog-to-Digital Converters with predictive auto-ranging

Floating-Point Analog-to-Digital Converters (FPADC) acquire both the mantissa and the exponent of sampled signals. Most of the FP-ADCs implement, either non-uniform quantization structures, or two step conversion techniques (firstly the exponent is gotten and then the mantissa is acquired). The first FP-ADC solution presents high conversion rates and reduced precision, while the second one preserves the precision of uniform ADCs but the conversion time is doubled. To minimize the acquisition time, an FP-ADC architecture was conceived (1999); it consists of two uniform converters that operate in parallel: one gets the exponent, while the other one, equipped with a programmable input amplifier acquires the mantissa. The current gain of the programmable input amplifier is based on the last exponent acquisition. This paper presents several techniques designed to improve the FP-ADC range prediction, to further minimize the conversion time, while increasing the SNR of the quantized signals. A polynomial extrapolation method for short term forecasting is considered. The SNR of a quantized signal that is acquired with the proposed FP-ADC is several orders of magnitude higher than that provided by a fix-point ADC, while the conversion time is shorter then those of other FP-ADCs.