Analog-to-digital converters (ADCs) are key components of digital signal processing. Classical samplers in this framework are controlled by a global clock. At high sampling rates, clocks are expensive and power-hungry, thus increasing the cost and energy consumption of ADCs. It is, therefore, desirable to sample using a clock-less ADC at the lowest possible rate. An integrate-and-fire time-encoding machine (IF-TEM) is a timebased power-efficient asynchronous design that is not synced to a global clock. Finite-rate-of-innovation (FRI) signals, ubiquitous in various applications, have fewer degrees of freedom than the signal's Nyquist rate, enabling sub-Nyquist sampling signal models. This work proposes a power-efficient IF-TEM ADC architecture and demonstrates sub-Nyquist sampling and FRI signal recovery. Using an IF-TEM, we implement in hardware the first sub-Nyquist time-based sampler. We offer a feasible approach for accurately estimating the FRI parameters from IF-TEM data. The suggested hardware and reconstruction approach retrieves FRI parameters with an error of up to -25dB while operating at rates approximately 10 times lower than the Nyquist rate, paving the way to low-power ADC architectures.Index Terms-Brain-inspired computing, analog-to-digital conversion (ADC), time-based sampling hardware, integrate and fire TEM (IF-TEM), sub-Nyquist sampling, finite-rate-of innovation (FRI) signals.

No abstract

Key parameters of analog‐to‐digital converters (ADCs) are their sampling rate and dynamic range. Power consumption and cost of an ADC are directly proportional to the sampling rate; hence, it is desirable to keep it as low as possible. The dynamic range of an ADC also plays an important role, and ideally, it should be greater than the signal's; otherwise, the signal will be clipped. To avoid clipping, modulo folding can be used before sampling, followed by an unfolding algorithm to recover the true signal. Here, the authors present a modulo hardware prototype that can be used before sampling to avoid clipping. The authors’ modulo hardware operates prior to the sampling mechanism and can fold higher frequency signals compared to existing hardware. The authors present a detailed design of the hardware and also address key issues that arise during implementation. In terms of applications, the authors show the reconstruction of finite‐rate‐of‐innovation signals, which are beyond the dynamic range of the ADC. The authors’ system operates at six times below the Nyquist rate of the signal and can accommodate eight times larger signals than the ADC's dynamic range.