System-level Techniques for Analog Performance Enhancement

All electronic circuits and systems perform two basic functions: signal generation and detection. Hence their performance or resolution are evaluated by the accuracy signal is handled at a certain speed. Therefore, the objective of all analog designs is to meet specified accuracy and speed requirements with constraints such as supply voltage, power, noise, or signal swing over process, voltage, and temperature variations. As CMOS is scaled down to the nanometer range, a new analog design style that relies on switching techniques has emerged and gained momentum. Fundamental concepts on key standard switching circuits such as comparator, latch, dynamic amplifier, and sampled-data circuit are elaborated in terms of transient noise, distortion, hysteresis, metastability, jitter, BER, and DC wandering.

Analog performance has been improved by careful design, new process, or feedback. Feedback is the only systematic way to enhance analog performance such as linearity, signal range, bandwidth, and impedance. Series (voltage) feedback increases linear voltage range while shunt (current) feedback increases linear current range. Also series feedback raises impedance while shunt feedback lowers impedance. Low impedance is for broad-banding while high impedance is for buffering. Feedback can be applied at any local to global levels, in continuous-time or discrete-time modes, and in broadband or DC servo applications. All feedback circuits from local to global and from series to shunt are handled consistently using the loop gain and phase margin concepts.

There are two types of circuits operating at almost DC. They are power supplies and sensor networks, in particular human body sensors for medical instrumentations. The former requires very low output impedance for transient uses while the latter needs narrow bandwidth. All electronic circuits need to draw energy from constant DC power sources for proper operations. Power supplies perform two basic functions of DC–DC conversion and voltage regulation. Voltage regulator is to reduce ripple in the supply output, but consumes power due to the voltage drop in the pass transistor. In recent years, switching DC–DC converters are commonly used for high efficiency together with low-dropout (LDO) regulators. Almost DC sensing circuits are for high-impedance instrumentations such as voltmeters, which suffer classic problems such as DC wander and motion artifact due to capacitive coupling. Based on the basic energy conservation law for charge and flux storage devices, the power efficiencies of DC–DC converter and LDO are redefined, and the transient distortion of LDO outputs is addressed using their output impedances. The DC wandering and motion artifact problems in almost-DC body sensor circuits are also clearly defined.

Data converters perform two functions: Data acquisition and data distribution. The former is to acquire digital data from analog channels for digital signal processing, and the latter is to distribute the processed digital data back to the analog channels. Two essential elements for these tasks are analog-to-digital converter (ADC) and digital-to-analog converter (DAC). There are various types of converters with a wide range of sampling rates and resolutions. Data converters help to overcome the flaws of switched-capacitor analog sampled-data processing, and make digital signal processing possible. Since it is not affected by analog imperfections, digital signal processing has replaced analog signal processing. Accuracy of the data converter is limited by magnitude and transient errors in the sampling and signal generation. The magnitude error is a static DC error that can be trimmed or calibrated while the transient error is a dynamic error that needs only to be controlled precisely. Basic circuit concepts for ADC and DAC are fully covered with special emphasis on their high-resolution implementations. In particular, the pros and cons of the successive approximation register (SAR) architecture are extensively discussed and compared to the oversampling continuous-time delta-sigma modulator (CT DSM).

The elusive goal of the lossless charge transfer from one circuit node to another has been the main focus of analog sampled-data processing starting from the bucket brigade and CCD to switched-capacitor circuits. Opamp with capacitive feedback is the most accurate analog component that can sample and amplify signal with highest accuracy. Their voltage transfer accuracy is solely dependent upon the DC gain and nonlinearity of the opamp. The lossless voltage transfer regardless of the opamp gain and nonlinearity error is achievable by eliminating it right from its source in the analog domain. When applied to the pipelined ADC, the linearity performance can be enhanced by adaptively cancelling them based on the global zero-forcing LMS feedback. Two featured circuit concepts can be incorporated to implement error-free switched-capacitor amplifiers. An all-analog feedback self-trimming mechanism is shown to solve the classic opamp gain and nonlinearity problem by employing a single-parameter adaptation scheme.

Old RF single-medium transceivers are mostly narrow-banded, and filtering at RF and IF helped to realize discrete RF systems. However, in modern integrated digital wireless systems, the lack of RF and IF filters heavily taxes on the performance of RF transceivers. In particular, channels with broadband spread spectrum and strong nearby blockers demand analog RF processing be extremely linear, and channel filtering highly selective. Direct frequency conversion to and from DC (Zero-IF) or low frequency (Low-IF) is the mathematical frequency translation using complex local (LO) carrier. On the transmitter (TX) side, key design factors are carrier leak, injection locking, out-of-band emission, spectral regrowth, and power amplifier (PA) efficiency. The carrier is suppressed using four-quadrant mixers, and the injection locking is avoided by dual conversion. The out-of-band emission is reduced below the spectral mask using RF filters or harmonic mixers, and the spectral regrowth issue doesn’t exist with low IM3. The PA efficiency is also traded for linearity. On the receiver (RX) side, there are self-mixing, blocker, offset, harmonic mixing, and image issues. The self-mixing with the local carrier is avoided using low-IF or dual-conversion architectures, and the blocker is rejected by digital channel filters with wide dynamic range front-ends. The offset in the zero-IF receiver is eliminated using the feedback that is active only during the data packet period. The low-IF receiver exhibits no offset problem. All issues related to the harmonic mixing and image problems in direct down-conversion receivers are discussed exhaustively using a generic complex frequency translation concept.