Online Memory-Constrained Frequency Estimation for Low-Resolution Non-Linear ADCs

Low-resolution analog-to-digital converters (ADCs) are prevalent in both low-cost applications where memory is highly constrained and in wideband spectrum analyzers and receivers with highly constrained latency. In both applications it is desirable to implement a fast, memory-efficient frequency estimator that operates on the output digital samples, whether for post-correction indexing or estimation of the location of spectral power in the ADC bandwidth. However, such ADCs are susceptible to non-linearities due to manufacturing tolerances and high-speed operation. We present a method for fast frequency estimation via low-resolution quantization and table look-up with table entry estimates analytically optimized for ADC-specific non-linearity patterns, as well as a zero crossing-based indexing method with superior memory efficiency for slightly higher latency and implementation complexity. The accuracy of these estimators is compared for various resolutions, window lengths, and evaluation SNRs, as well as in the presence of ADC non-linearity. Our results indicate that the zero-crossing estimator far outperforms the conventional look-up table estimator in every SNR range for both memory and accuracy.