February 2, 1998

AGC disciplines RF and fiber signals so they "ain't misbehavin'"

BILL SCHWEBER, TECHNICAL EDITOR

You may need AGC to keep wayward RF, optical, and video signals--which swing over a wide dynamic range--within acceptable bounds. By considering key specifications and techniques, you can get optimum AGC results.

Digital signals were supposed to minimize the nasty problem of information-bearing signals that have undesirably wide dynamic ranges. Sure, digital signals might have some additive noise, but at least they generally stick close to their nominal levels. Although analog signals, such as RF waveforms, would still have varying power levels, the ability of digital circuitry to retrieve and restore pristine signal levels would make dynamic-range problems much less common.

But the real analog world didn't succumb to that forecast. Wireless systems in new and widespread applications have increased the demands on analog circuitry, which must capture the usually minuscule front-end signal and make it viable--an increasing challenge when the signal strength fluctuates because of rainfall attenuation, tree absorption, and walls and openings in buildings. Fiber-optic signals en-dure loss and variability from weakening emitters, from fiber variations, and through switches. Video-signal sensors must operate under light conditions from near dark to bright sunlight.

The AGC circuit, a classic analog function that engineers have used since the 1930s in basic radio circuitry, has increased in importance. The function of an AGC is simple: to keep the maximum and minimum signal voltages or power levels within desired boundaries despite inherent widespread variations and to do it by automatically controlling gain in a closed-loop control function, based on the input-signal strength.

The more things change...

The underlying problem is this: In many applications, it's the modulated-signal envelope, not the absolute re-ceived-signal value, that conveys information, and this signal-envelope span is too uncontrolled in your circuit to properly analyze. The wide range either saturates the next stage, or, if you attenuate the signal to provide head room, falls below an acceptable noise level.

For example, 1 bit equals 1.5µV in a 16-bit ADC with a 0 to 1V input span. If your design's average signal drops below a few least significant bits, it loses most of its ability to characterize the actual signal with high resolution and likely loses the signal in noise. The same situation is true if the signal's average value is near the high end of the span. Either way, your actual resolution is not one part in 2¹⁶, but only a few bits. And if the signal drops too low or balloons too high, you lose all potential resolution.

Look at this situation from another perspective. A converter's dynamic range, in decibels, is six times the number of bits, so a 10-bit converter has 60 dB of range, and a 16-bit device has 96-dB resolution. If the carrier average-signal-strength swing is 50 to 80 dB, which is common in many applications, you lose most or all of the head room you need to distinguish the information embedded within that carrier.

Even if you don't use an ADC to extract data from a signal, variations in nominal signal strength affect analog-signal processing. In a fiber-optic system, you may see 30 dB of power variation at the photodetector output. When picoseconds are critical, such as in systems with rates of more than 100 Mbps, the timing-extraction circuit is prone to more jitter when the signal's amplitude variations prevent a continuously optimal match be-tween anticipated amplitude and recovery-circuitry design (Figure 1). Without gain control, the zero crossing of the full-amplitude signal shifts from that of the lower amplitude signal be-cause of distortion. Using an AGC circuit levels the amplitudes, thus reducing the recovered timing jitter and the system bit-error rate.

The basic AGC circuit neatly solves these problems without active intervention by the rest of your system (Figure 2). The circuit uses a measure of the input-signal strength to control the gain of a voltage-controlled amplifier (VCA). This measure of strength can be input-signal voltage amplitude or signal energy or power as derived from the voltage. Although the AGC function has one input and one output, some AGCs also provide access to the control-voltage output of the lowpass filter, which drives the VCA. You can use this gain-control voltage as a received-signal-strength indicator, which some applications need to adjust other circuit parameters.

The lowpass filter is critical to overall AGC dynamics, just as it is in a PLL (Reference 1). You must balance any initial attractiveness of fast response against transient response and even loop instability. For example, you can use a peak detector rather than a root-mean-square (rms) detector to adjust the gain when the input crosses a threshold. In making this substitution, however, you allow impulse-noise spikes to cause sudden  decreases in gain; these decreases are often undesirable if the actual signal strength is otherwise low and the gain loop takes time to recover.

Note that many engineers often use the terms "AGC" and "VGA" (variable-gain amplifier) interchangeably, be-cause AGCs use VGAs as cores. Although this substitution is not strictly correct, the difference between the two is the addition of the loop-closing path to the VGA. Also, before you assume that the AGC function is the solution to your range-matching problem, look at other functional relationships that may be appropriate (see box "Analyze the relationship").

AGC circuits face varying signal conditions. For RF and fiber signals, gain-control adjustment range may be 40 to 100 dB; for analog video signals, the required span is typically just a 4- or 5-to-1 range. Accuracy requirements differ as well: Gain leveling with 5 to 20% accuracy usually satisfies wide-ranging RF and fiber signals, but video signals often need leveling to less than one IRE (Institute of Radio Engineers) unit, or approximately 0.5%.

Loop response is another AGC issue. A short time constant causes gain changes in response to the modulating signals or data bits themselves. Thus, a faster response is often less desirable than a slower one that allows the circuit to assess signal strength over many carrier cycles or bit periods. You should set the initial loop time constant in terms of the expected rate of signal-strength changes and not the embedded data rate or modulating-signal bandwidth. Although the best time constant depends on the application, start with values of approximately 10 µsec for RF and fiber-optic links and 60 to 100 µsec for video (equal to a horizontal-sync frame).

Two other dynamic specifications of an AGC function need your attention. First, the gain-control response time tells you how fast the gain setting of the AGC slews from one value to the next when directed by the gain-control signal. This response time must be compatible with the loop response time you choose. Further, unless you can ensure that your input signal is well-behaved and will stay within a maximum value, check the overload response of the AGC to see how long it takes to recover from a saturated state. During saturation, while the AGC is temporarily immobile, your signal path is no longer linear. Thus, the consequences of overload may affect your system design and signal-processing algorithms.

Don't forget that, at its core, an AGC circuit provides an amplifier function. This fact means that you need to look at traditional amplifier considerations, such as bandwidth, linearity, distortion, noise, input conditions, and output drive, but with an added layer of subtlety because of gain changes. The signal in your application determines the needed amplifier bandwidth: about 5 MHz for composite video but 30 MHz for high-definition TV. Note that bandwidth changes with gain settings for voltage-feedback amplifiers, so optimizing gain range vs bandwidth involves careful consideration of the trade-offs in your system.

Be careful with both harmonic and intermodulation distortion, especially at higher frequencies. Typical distortion-value requirements are 30 to 40 dB or more below the input-carrier amplitude. For video applications, keep minimal differential gain and phase errors at 0.1% and 0.01°, respectively; some applications have even tighter tolerance requirements.

Evaluating noise effects in an AGC  circuit is more difficult than it is with a conventional fixed-gain amplifier. Although your system design can sometimes compensate for or accommodate a constant rms noise magnitude, an AGC has varying gain, and, thus, its output noise level varies as well. Noise also tends to increase with bandwidth, so minimize bandwidth, commensurate with your dynamic specifications.

Another noise source in wideband AGC applications is high-frequency feedthrough. Even if you set the amplifier for maximum attenuation, some of the signal feeds through because of internal IC capacitance, board layout, and power-supply signal paths. Think of this feedthrough as a noise floor in your error-budget analysis.

The output of an AGC typically has a high-impedance drive and needs a good buffer to drive subsequent stages and to possibly scale the output magnitude. Make sure that the amplifier specifications for bandwidth, distortion, and noise don't defeat the AGC and that you have sufficient current drive and output swing. As with all high-speed amplifier designs, layout is critical. Bypassing, ground planes, and input- and output-signal paths all need the attention you would give a conventional amplifier.

Don't expect the AGC to do it all. Taking a microvolt input and making it large enough while handling larger-than-desired signals--and achieving these tasks with low distortion--are a lot to expect. Most AGCs have a nominal output magnitude at which their distortion is at a minimum, such as 100 mV. You may need to use a low-noise preamplifier stage ahead of the AGC circuit to remove some of the gain burden from the AGC and to improve the SNR. The level at which the AGC operation asserts and thus adds gain is also critical, because this level is where you have additional noise on that small, fragile input signal.

Look at both the minimum detectable signal, which circuit noise primarily governs, and the overload head room and consequences of amplifier saturation. You may even want to explore ganged AGCs, which can solve a gain-partitioning problem, give more freedom in selecting optimal operating points, and provide two independent time constants.

AGC ICs and functional blocks are available from both broad-line op-amp vendors and sources that specialize in AGCs for specific applications (Table 1). Be prepared to spend a lot of time studying their  data sheets. These data sheets have many graphs defining performance under various operating conditions, and the AGC data sheets have even more graphs because of AGCs' inherent gain variability. The data sheets characterize various aspects of AGC performance at nominal supply, such as 3.6V, as well   as at lower voltages--2.4V, for example. The sheets also present temperature-related performance factors, although these factors are often less critical than other parameters because of the closed-loop nature of the AGC action, which tends to compensate for some temperature-induced variations.

Elantec, for example, offers a VGA with maximum gain fixed at +2 and 70-MHz signal bandwidth, targeting input-signal spans of ±2V. The EL-4451C has differential inputs and includes an output amplifier, and its output slews at 400V/µsec. If you prefer a multiplier, the EL2082C current-mode multiplier has 150-MHz large- and small-signal bandwidths, with 46 dB of calibrated gain-control range. Specified for video applications, this multiplier has 0.15% differential-gain error and 0.05° differential-phase error under NTSC test conditions.

If you prefer to adjust your gain in decibels using a linear signal, the AD603 from Analog Devices is worth considering (see box "When bad things happen to good signals"). This calibrated VGA for use in RF/IF AGCs lets you select gain range of ­11 to +31 dB with a 90-MHz bandwidth or 9 to 51 dB with a 9-MHz bandwidth.You can also select intermediate gain ranges. The VGA is scaled at 25 mV/dB and needs a 0 to 1V signal to control gain over the 40-dB range. Gain-control re-sponse time is less than 1 µsec for the full range; input-noise spectral density is 1.3 nV/square root(Hz).

The CLC5523 VGA from the Comlinear Group at National Semiconductor provides 250-MHz bandwidth (75 MHz with 0.2-dB gain flatness) and 60-dB gain range, and its gain-control response rate is 4 dB/nsec. You can set the maximum gain from 2 to 100 using two resistors and control the gain level using a 0 to 2V signal. The CLC-5523's data sheet represents a typical AGC and VGA, and provides a graph of frequency response with changes in gain-control voltage (Figure 3).

For fiber-optic applications, Vitesse offers two amplifiers with integral AGC. The VSC7902 for 155-Mbps SONET/SDH links and the VSC7911 for 622-Mbps links provide broadband amplification with a single-ended input and output; differential-output versions are also available (Figure 4). You set the AGC loop time constant using an external capacitor, with a 470-pF unit yielding a 33-µsec result. Noise is critical in high-speed fiber applications, and the input-noise-current density of the VSC7911 is 2.8 pA/square root(Hz).

Don't spend your time reinventing the signal path and AGC implementation if you can avoid it. Vendors often offer evaluation boards that have worked out the layout and application subtleties that you'll have trouble rediscovering until it is often late in your project's design cycle. For example, Qualcomm offers a pair of dual-mode  CDMA/FM AGCs for cell-phone receiving and transmitting paths. The receiving-channel Q5500 for 10- to 300-MHz operation has a 90-dB gain-control range, and the transmitting-channel Q5505, for 10- to 250-MHz operation, has a control range of 84 dB. In addition, Qualcomm offers a separate evaluation board for each IC that lets you assess performance with both single-ended and differential configurations using a jumper that selects signal-path mode (Figure 5).

The relentless march of analog integration onto one IC has affected AGCs, too. You may find that an embedded AGC, as part of a larger functional device, is sufficient for an application and thus eliminates the need to provide a separate VGA function. For example, the MAX2102 from Maxim tunes L-band signals from 950 to 2150 MHz and converts them to baseband. In addition to the requisite low-noise amplifier, the IC includes a front end with a 50-dB VGA for carrier signals from ­69 dBm to ­19 dBm.

By embedding the AGC into the larger, application-specific IC, the vendor can also make trade-offs rather than provide a device that must be superior in all dimensions. Knowing the application, the vendor may, for example, concentrate on minimizing distortion but give up some unnecessary gain-control range.

Reference

1. Schweber, Bill, "PLL synthesizers make channel-hopping swift and sure," EDN, March 14, 1997, pg 51.

Acknowledgments

Thanks to Kenneth Fields of Elantec Semiconductor Inc; Ray Milano, Robert Deming, and Bala Mayampurath of Vitesse Semiconductor Inc; Jonathan King of Qualcomm Inc; and Debbie Brandenburg of National Semiconductor Corp for their insight and comments.

| EDN Access | Feedback | Table of Contents |