扩展电流反馈放大器的可用性(2)

晶闸管电源

Amplifiers: Op Amps

Texas Instruments Incorporated

introduce errors into the system due to

the bias current and the dynamic signalcurrent flowing through this impedance;but these effects are reasonably small as

long as the impedance is minimized.

Adding impedance Z can affect inputoffset voltage due to the dc input biascurrent, which is typically 1 to 10 µA,

multiplied by the impedance Z. Thisresulting voltage gets multiplied by thenoise gain of the circuit. Additionally, whena signal appears at the output, the CFBamplifier (as the name implies) relies on

an error current flowing through theinvertingnode through the impedance Z,producing a signal error. However, sincethe transimpedance of most CFB ampli-fiers is well over 100 k and sometimes ashigh as several megohms, this error is also

minimized if the impedance is kept low. The

drift of this circuit now also relies on the

temperature characteristics of impedance

Z and should not be used as a precisionamplifier; but most CFB amplifiers are notused as precision amplifiers anyway due to

stated previously. This shows that there is a reasonably widetheir inherent topology limitations. Overall, these issues

range of acceptable values for Z and does not imply that theare minimal and, for most systems, can be effectively

selection for Z is highly critical. Figure 3 also illustrates aignored in favor of the CFB amplifier’s advantages as

common trait for current-feedback amplifiers—as the feed-previously stated.

backimpedance is decreased, the peaking will increase. If

Testing with different Z valuesthe impedance is too low, there is a good chance that theThe easiest way to see if the circuit is stable is to use acircuit will become unstable and oscillate, as illustrated bynetwork analyzer frequency sweep. Instability can typical-the response when Z = 200 .

ly be seen as sharp rises in the frequency response at the

Output noiseamplifier’s bandwidth limitations. If the peaking is smooth,

One element that may be very important in a system is theor there is no peak, then the amplifier should be stable.

output noise. Adding a resistance in the manner discussedFigure 3 shows the frequency response of the system with

only makes the output noise worse. The inverting currentdifferent values of resistors for the variable Z.

noise of the amplifier goes through the resistance at Z andThe response of the THS4012 is also shown for reference

creates a voltage noise. This noise then becomes multipliedto easily compare the performance of the two systems. It

by the circuit’s gain, which is frequency-dependent.is interesting that no matter what resistance is used for Z,

For a CFB amplifier, the inverting current noise is typi-the responses below 20 MHz look identical to each other.

cally the highest noise component of the amplifier. AlthoughThis is the ultimate goal of this configuration—no differ-the CFB amplifier voltage noise is inherently very low, ences in signal performance. For the stability part of the

——

typically less than 3 nV/√Hz, the inverting currentnoise ofcircuit, the area above 20 MHz must be examined.

——

most CFB amplifiers is generally around 15 to 20 pA/√Hz.Examining the circuits in Figures 1 and 2 shows us that

The noninverting current noise is only noticeable if thethe feedback impedance is dictated by the capacitor CF.

source impedance is high. Using a 50- environment Above 20 MHz, this impedance is very small—essentially

minimizes the noninverting current noise.creating a short from the output to the summing node. This

The THS3112 was designed to have very low noise. Theconfiguration is commonly referred to as a unity buffer with

——4voltage noise is 2.2 nV/√Hz, the noninverting current noisethe signal gain set to 1. The data sheet for the THS3112

——

is 2.9 pA/√Hz, and the critical inverting current noise is arecommends that, in a gain of +1 under the circuit condi-——

low 10.8 pA/√Hz. However, multiplying the inverting currenttions utilized, the feedback resistance be 1 k . Thus, it is

noise by 1 k and then multiplying by the gain can aloneno surprise to see that when Z = 1 k , the response looks

——

produce a very substantial output noise of about 54 nV/√Hzvery smooth and well behaved, indicating a very stable

in the pass band. To quantify the output noise of the system,system. However, when Z = 681 , the response also looks

the circuits shown in Figures 1 and 2 were tested for outputvery reasonable and helps minimize the potential issues