No abstract

This paper presents a low-power, time-based, compressive sampling architecture for analog-to-digital conversion. A random pulse-position-modulation (PPM) analog-to-digital conversion (ADC) architecture is proposed. A prototype 9-bit random PPM ADC incorporating a pseudo-random sampling scheme is implemented as proof of concept. This approach leverages the energy efficiency of time-based processing. The use of sampling techniques that exploit signal compressibility leads to further improvements in efficiency. The random PPM (pulse-position-modulation) ADC employs compressive sampling techniques to efficiently sample at sub-Nyquist rates. The sub-sampled signal is recovered using a reconstruction algorithm, which is tailored for practical hardware implementation. We develop a theoretical analysis of the hardware architecture and the reconstruction algorithm. Measurements of a prototype random PPM ADC and simulation, demonstrate this theory. The prototype successfully demonstrates a 90% reduction in sampling rate compared to the Nyquist rate for input signals that are 3% sparse in frequency domain.

No abstract

Neurons as Time Encoding Machines (TEMs) have been proposed to capture the information present in sensory stimuli and to encode it into spike trains [1], [2], [3]. These neurons, however, produce spikes at firing rates above Nyquist, which is usually much higher than the amount of information actually present in stimuli. We propose a low-rate spiking neuron which exploits the sparsity or compressibility present in natural signals to produce spikes at a firing rate proportional to the amount of information present in the signal rather than its duration. We consider the IAF (Integrateand-Fire) neuron model, provide appropriate modifications to convert it into a low-rate encoder and develop an algorithm for reconstructing the input stimulus from the low-rate spike trains. Our simulations with frequency-sparse signals demonstrate the superior performance of the Low-Rate IAF neuron operating at a sub-Nyquist rate, when compared with IAF neurons proposed earlier, which operate at and above Nyquist rates.

No abstract