|
|
|
|
|
|
24 July 2008
|
Building Blocks
Gaining Control
AGC, another familiar communication system piece, gets the usual block-diagram level of attention in most system descriptions. In this month’s column, we’ll dissect this heretofore unrecognized item.
By Rob Howald
I
love the block-diagram descriptions that I regularly encounter during my daily grind. There are nice labels for all pieces of the input and output devices of a communication link. And, invariably, there is an HFC network cloud between the blocks. The efforts of my entire business unit focus on that cloud sandwiched between a couple of blocks. This type of over-simplification occurs within communication texts as well. In the receiver of many communication system diagrams, we find the automatic gain control
(AGC) block. The block may also be labeled as the automatic level control (ALC). Although the AGC often appears as one of those zero-impairment magical squares in a block diagram, it can actually be a tricky little function if not handled properly.
A block diagram of fundamental AGC functionality is shown in
Figure 1
. It’s shown in the ana-log sense because, like most processing and analysis operations, understanding it with our analog minds is easier
in the analog world.
For engineers who like playing with both circuits and analysis, AGC design is one of those items with a little of everything, when performed in the analog domain. While the circuit in question is often an RF or microwave circuit, it still requires a knowledge of cascade analysis, analog design, control systems, and feedback analysis.
The function
The goal of an AGC operation (or AGC loop, so-called because it is a feedback, or “loop,” system) is to
develop a constant power level through the subsequent stages. At some point in a receiver, a known signal level must be counted on to economically and reliably design processing stages that operate near optimum in terms of drive and output level. For subsequent stages of analog processing, this optimal point is determined by the relationship of signal level to noise or distortion limits. Exactly where the level should be within this dynamic range window depends on the specific cascade and where the AGC sits
within it. This is where the designer’s cascade skills come into play. The function will typically sit between a couple of other modules (such as a low noise amplifier [LNA] and a mixer/downconverter stage).
For an analog AGC loop that is ahead of an analog-to-digital converter (ADC), the conversion that takes place should do so with the required conversion signal-to-quantization noise ratio (SQNR). Never one to miss a chance to shamelessly plug a prior effort — a two-part
“Building Blocks” series on analog-to-digital conversion and its related terminology is available from the
Communication Systems Design
archives (“Converter Concepts and Specs: Parts I & II,”
Communication Systems Design
,
June 1998
&
July 1998
). In general, an ADC’s discrete nature assigns a binary value to each continuous amplitude sample. Only a finite number of choices exist with a digital word. In fact, two
voltages that are very close may be represented by the same digital word. The average amount of error between the actual signal and the discrete representation is described as quantization noise. Quantization noise is an absolute number that is strictly a function of the input analog range and the number of converter bits. Because of this fixed average error, the way to ensure a high post-conversion SNR is to make sure the converted word is of high amplitude. This translates to hitting the ADC at its input
as close to the desired maximum as possible, maximizing the SQNR. A happy (but not always practical) goal for the converted SQNR is to not contribute to link SNR performance. In other words, if the SQNR is always 10 dB better than the SNR created by the rest of the link, then its contribution to degradation can be considered negligible. Of course, assurance of this occurs at the expense of the processing throughput required for high resolution. An ADC, much like a subsequent analog cascade, must be
driven at a level that optimizes (in some sense) its processing performance. An AGC loop provides this control.
In digital interfaces, an AGC section provides sample magnitudes to help optimize a subsequent receiver’s functional cascade. Ultimately, controlled-amplitude sample values are needed to provide symbol detection for a digital communication stream. In digit land, special constraints on word size and the number of bits needed to represent values exist, and the AGC function provides an
important normalization to ensure reliable and mathematically sound bit manipulation.
While this may explain what we are attempting to do, we have sidestepped one obvious item — why we need to do it at all. Signal-level amplitudes that reach a receiver can vary wildly for a number of reasons, especially on channels that rely on RF propagation. The following signals can be affected by a multitude of factors: satellites (weather, clouds, rain); wireless networks (distance, fading); microwave
point-to-point; MMDS; and LMDS (the same obstacle course of free-space impairments). This brief list of the atrocities endured by some of these signals on propagation paths can result in the need for wide dynamic ranges. I sense readers in suspense, mouthing, “How wide?” I still can’t answer that completely, except to say that in my experience, I’ve seen designs asking for as much as 70 dB, designs which need only 10 dB, and a slew of designs looking for something in the 30- to 40-dB range.
I’d particularly expect the 70-dB designs to be prevalent among the radar sect, where designers are slaves to unearthing every half-dB of dynamic range. In any event, the required dynamic range translates into the necessary number of variable-gain amplification stages. In general, a discrete device single stage can cover somewhere between 20 and 40 dB, depending on frequency of operation and implementation (which can be either variable attenuation or true variable amplification).
Detection and
threshold comparison
An easy way to conceptualize an AGC loop (and the fastest way to get one operating) is to think of the entire unit as a DC operational amplifier circuit rather than an RF-level control circuit. Do you remember — this could be painful, perhaps bringing out some long-suppressed horrific memory of an exam — operational amplifier properties? If you occasionally dabble in analog design, you take one important property for granted at every turn — the voltage on the two
input terminals that represent internal differential amplifier terminals is the same. This property is used to generate all of the basic circuits — summers, difference amplifiers, integrators, filters, and so on. Thinking of an op amp as the center of the circuit is also a simple way to look at an AGC loop.
Consider that one terminal (the positive terminal) has a voltage proportional to the gain value of the desired AGC loop in steady state (see
Figure 2
). For comparison to this threshold, we need a measure of the RF amplitude we are trying to control. This is the function of an RF detector. This component takes an RF (or AC) input, and delivers a DC output proportional to the input voltage (or power, depending on the mode of operation). The device may have the fanciest names applied to a choke, diode, and resistor-capacitor (RC) combination, cleverly obfuscating the components within and allowing a thriving marketplace of RF detector manufacturers. A
detector does the following: It rectifies the AC waveform, creating a DC component. The rectified waveform is filtered by a parallel RC combination. The choke is often used to provide a bias current path through the diode along with the resistor, while not effecting AC, so that all of the energy impinges on the diode. Depending on the diode type (frequency of operation, sensitivity, and so on), a bias current improves performance. In any event, an RF input results in a DC output. Now put this detector at the
end of an RF cascade, with variable gain controlled by a voltage-controlled attenuation that decreases attenuation with increasing voltage. The detector’s gain-control port is fed by the filtered output of the op amp.
Plug this DC component into the negative op amp terminal. If positive exceeds negative, then the threshold is higher, which is another way of saying the signal is too low or attenuated too much. The op amp output may head up toward higher voltage, but the output terminal is
low-pass filtered (averaged), and slowly ramps upward. The slowly increasing voltage is driving the voltage-controlled attenuator towards less attenuation. The signal and the detector increase, moving the voltage at the negative terminal towards the positive terminal. Eventually an equilibrium is achieved. This is the basic operation of detection and threshold comparison.
Other important matters
The fun begins for RF designs once we take that step towards real implementation. This is always the
case when dealing with what seems like RF voodoo at times, and it’s no different here. But much like deliberate, thorough RF design, there are fewer surprises when the design is thought through ahead of time. Here are some of the caveats of AGC design.
-
Range.
The first impact of a wide range is simply to require more than a single variable stage. This is not a particularly big deal, and is, in fact, no deal at all if the same component is used in the same frequency range.
Simply driving the input with the same AGC control voltage does the trick. However, things can change when different frequency ranges are present, such as if the AGC is broken up into RF and IF, and/or if different control elements are used. Different frequency ranges may be necessitated by gain splitting or convenience. Next, it is necessary to carefully assess the regions over which the gain control curves operate. This assures that the coverage is obtained with the proper overlap of windows of operation.
A very large dynamic range presents a more intricate problem. For example, a 70-dB gain range in two stages that are physically and electrically close can be hazardous. More than likely, your next design review will be for this novel integrated AGC amplifier/ oscillator design. That much gain can easily be enough to create a positive feedback path at some frequency via DC paths, grounding nonidealities, or cavity propagation. Thus, separation of stages with other passive, lossy elements may be
necessary. This separation may encompass an RF-to-IF stage conversion that eliminates overlapping high-gain frequency ranges, which may sing against your will.
-
Fun with feedback.
AGC analysis and principles follow the same type of closed-loop feedback analysis that is used for phase-locked loops. This may or may not be a useful analogy to you, so think of it in terms of the prior op amp example. A control system consists of the comparison between the input signal and the reference
signal, from which an error signal is generated. This error signal is low-pass filtered (averaged), and used to control the input signal, using negative feedback, and moves towards a zero-error stable condition. The key words in this description that define everything that is important are “filtered” and “stable.” Because the system is a feedback mechanism, it must be analyzed for stability, or put another way, we have to make sure the feeding back of processed signal components does
not cause the loop to oscillate. The approach to assuring stability is through careful design of the low-pass filtering. A pure integration function or an integrator augmented with finite DC gain are both common. The analysis tools for stability are the dreaded bode plots — gain and phase margin, which is actually not that difficult once the cobwebs are cleared out. It’s all very straightforward to a first approximation. But what makes AGC loops especially entertaining are the not so
straightforward things:
-
Gain curves.
Voltage versus attenuation and voltage versus gain curves in amplifiers are nonlinear. The slopes on these curves relate to AGC loop bandwidths, which in turn relate to AGC dynamics. Dynamics relate to the AGC’s ability to capture a signal, as well as how it responds or doesn’t respond to modulation. It can get ugly if the window of operation for the variable elements is not controlled. In particular, the slopes of gain versus voltage
near the endpoints can be extremely shallow or steep. The slopes can be linearized at the expense of the added circuit complexity, and/or cost.
-
Distortion.
The variable gain devices (again) are notoriously unpredictable in distortion performance versus voltage, and their position within the cascade is important, so as not to impact the end-of-cascade distortion performance. Obviously, it is important to characterize variable gain devices across their range of operation.
- Stability
. All of the pretty stability analysis comes crumbling down when you turn it on in the lab and see a pair of friends on either side of the spectrum analyzer’s main signal, indicating some loop oscillation. Chances are, you’ve been bitten by the parasitic capacitance bug, which may lurk within op amps, control ports to the variable gain elements, and in unsuspecting RF filter skirts within the RF cascade of the AGC chain of elements. If an added cap puts an added pole in the wrong
place, phase margin will evaporate.
- Modulation.
Continuous wave signals are easy to control, but the real ones carry information. In some cases, the information is in the amplitude, and in some cases it’s not, but the amplitude may still vary due to noise or channel filtering. The AGC bandwidth must be a compromise of how rapidly you want to keep up with drifting signal power, but not at the expense of tracking noise or modulation variation. The latter effectively strips off the
variation that a demodulator needs to extract the actual information. This is usually not a difficult trade-off to make, as signal variations are much slower than modulation variations, and the loop bandwidth can be low.
-
Noise figure.
AGC circuitry, especially of wide dynamic range, can seriously impact the receiver’s noise figure (NF) performance. It’s important to note which end of the gain the receiver’s NF spec takes into account. While it may sound like a daunting
problem, it’s generally the case that the lowest NF is needed for the lowest signal level, or highest gain condition. When the signal is stronger, attenuation is applied (assuming this implementation includes variable attenuators), and the NF will degrade. But the signal is stronger to begin with, so the NF degradation is often not a problem if it varies one-to-one or less with input signal level.
Tracking, acquisition, and synchronous AGC (which combines phase-locked detection and gain control
for optimal receiver performance, typically in very low SNR situations) are other interesting topics concerning AGC — which, due to space limitations, we cannot discuss in this month’s column. Then, of course, there is the digital receiver version and the associated numerical delights of that implementation. Hopefully, this discussion will inspire you to put down that Stephen King novel tonight or forego Monday Night Football, and dip into the archives for some light educational reading.
Rob Howald is the manager of systems engineering in the transmission network systems group at General Instrument in Hatboro, PA. He has a BSEE and an MSEE from Villanova University, and received his PhD from Drexel University. He can be reached at
rhowald@gi.com.
Return to the
Table of Contents
|
|
|
|
|
|
|
|
|
Puzzled by a network processing design issue?