Synthesis of Log-Domain Filters from First-Order Building Blocks

Abstract

Abstract. Log-domain ®lters are an important class of current-mode circuits having large-signal linearity and increased tuning range over voltage-mode ®lter circuits of similar complexity. In this paper we describe synthesis of a single-ended, ®rst-order ®lter circuit from static and dynamic translinear circuit principles, and show how higher-order ®lters can be easily constructed from the ®rst-order building block. We address additional issues related to low-frequency (audio-frequency) ®lter design and present results measured from test circuits and a complete 15-channel ®lterbank system fabricated in 2 mm and 1.2 mm BiCMOS processes.

A subthreshold aVLSI implementation of the Izhikevich simple neuron model

“…The circuit design utilizes a four transistor log domain filter [8] as the building block with MOSFET's biased in the subthreshold region. The drain current for this region of operation, I D is an exponential function of the gate-source voltage V GS :…”

“…We will refer the interested reader to excellent treatises on the subject [8,9,10] and concentrate on the application of this theory to the simple model. The reader is introduced to the simple model in Section 2.…”

Analog VLSI Biophysical Neurons and Synapses With Programmable Membrane Channel Kinetics

“…The addition of the capacitor transforms the circuit from a translinear multiplier into a log-domain filter [59] that implements the desired first-order dynamics. The derivation is provided in Appendix A.…”

“…6 by implementing the kinetics in the channel gating variables by combining the and rate currents. The log-domain circuit [59] uses the dynamic translinear principle, exploiting the exponential current-voltage dependence of MOS transistors operating in the subthreshold region [56]. Drain currents are modeled as in gate voltage and source voltage relative to the bulk, where is the thermal voltage, and is the bulk back-gate effect factor [57].…”

Neuromorphic processor for real-time biosonar object detection

“…For training and test, the data is separated in equal parts according to alternating angles for each object. A parallel constant-É bandpass filterbank topology [12] is used for the cochlear frontend with 32 center frequencies linearly spaced from 15 to 150kHz. Object labels define 6 classes for support vector machine supervised learning.…”